145
Chapter 5
Rentier State Theory versus Rentier State Reality: Economic Benefits and Individual Political Behavior in Bahrain
More than simply give empirical evidence to a general political disagreement between Sunnis
and Shi‘is in Bahrain, the present chapter seeks to evaluate the specific theoretical arguments
elaborated thus far in explanation of the larger case of Bahrain, of the case of the dysfunctional
rentier state. As such, it proceeds deliberately in toe with the extant rentier framework in order
to show how far its predictions about individual political behavior in the Arab Gulf accords
with the reported actions and opinions of our real-life Gulf Arabs. Since these predictions
stem directly from the economic devices through which allocative regimes are pretended to
buy popular political quiescence, our own line of argument has offered an alternative account
of the function of these rentier mechanisms under conditions of ethnic division, a modified
theory of contemporary Gulf politics that explains the behavior of Gulf citizens and rulers in
terms of intra-state ethno-religious rivalry tied inextricably to the region’s furious geopolitics,
dominated by a near-hysterical fear of Iran and of Iranian-inspired Shi‘a emboldening. These
revised predictions we may restate as a series of hypotheses that both summarize our changes
to the standard rentier model and organize our quantitative analysis of the Bahrain mass survey.
The first two of these correspond to the supposed ability of rent-based states to co-opt
domestic political support through the judicious and liberal use of public employment. Yet
in the case of the ethnically-contested rentier state, we have argued, this course is curtailed
by the veritable exclusion of the Shi‘a from the largest state employers—the police, the armed
forces, and to a slightly lesser extent the intelligence services (one always needs infiltrators)—
and their disproportionate exclusion from the power ministries and other so-called “sensitive”
bureaucracies for fear of their treachery. And as for those Shi‘is who do find government
employment, we said further, they will be suffered to fill comparatively lower-level positions,
146
as per the incisive political cartoon of Chapter 3, which has the “Ministry of Sectarianism”
distributing “government jobs” to Bahrain’s Sunnis and “administrative positions” to the Shi‘a.
Thus we arrive at the following testable conclusions:
Hypothesis 1.1: Ethnic affiliation is a significant predictor of public-sector employment in Bahrain,
such employment being negatively associated with Shi‘i identification; and
Hypothesis 1.2: Among public-sector employees, Shi‘i ethnic membership is negatively associated
with occupational level. The second set of empirical predictions corresponds to our theoretical critique of the
other alleged rentier pacifying mechanism, the so-called “taxation effect” whereby citizens of
rentier regimes, not asked to pay taxes, in return are loath to make any political demands on
their leaders out of an apparent moral insistence on subject-ruler parity, an ethical qualm
with receiving while not giving. On this point, we said, rentier theorists may well be correct:
untaxed citizens may, all else being equal, demand fewer political goods than their taxpaying
counterparts. But far from implying a disinterested public, this statement serves only to rule
out a single possibility from among the many plausible drivers of popular political zeal, not
all of which will revolve around economics. In short, we said, the rallying cry of the American
Revolution need not coincide with that of the Islamic Revolution; in the Arab Gulf, ethnic
solidarity and rivalry offers a viable basis for political coordination. This argument suggests:
Hypothesis 2.1: Ethnic affiliation is equally significant as, or more significant than, economic well-
being in predicting individual political opinion and action in Bahrain; where
Hypothesis 2.1A: Sunni ethnicity is associated with less political action; and
Hypothesis 2.1B : Sunni ethnicity is associated with more positive political opinion; further,
Hypothesis 2.2: The strength of ethnic identity is a significant predictor of individual political
opinion and action in Bahrain; where
Hypothesis 2.2A: Among Shi‘is, stronger ethnic identity is associated with more negative
political opinion; while
Hypothesis 2.2B: Among Sunnis, stronger ethnic identity is associated with more positive
political opinion; and
Hypothesis 2.2C: Among Sunnis as well as Shi‘is, the strength of ethnic identity is positively
associated with political action.
147
The distinction between Hypotheses 2.1 and 2.2, then, is that while the former simply
posits a between-group discrepancy in survey responses—i.e., that Shi‘is will tend to report
being more politically-active and -interested; and will report less favorable opinions about
Bahrain’s government and general political situation—the latter hypothesis goes a step further
to predict that these effects are further augmented as the ethnic identity (defined later as
“religiosity”) of a respondent increases in strength. In other words, Hypothesis 2.2 posits that
the more strongly a Bahraini identifies as a Sunni or a Shi‘i, the stronger the respective effects
on his political opinions and actions. This argument entails specific, falsifiable predictions the
substantiation of which will lend considerable support to our larger thesis outlined in Chapter 2.
Particularly crucial in this regard are the subsidiary Hypotheses 2.2B and 2.2
C, which represent our
claim that ethnic-based political mobilization in Bahrain is not a phenomenon driven wholly
by Shi‘a but entails a countervailing force composed of ordinary Sunnis aimed at preventing
their rivals from gaining political power and influence. We should, accordingly, find that the
same measures of the strength of one’s ethnic identity (or “religiosity”) have a different—
indeed opposite—effect depending on whether one is Sunni or Shi‘i: for Shi‘is, greater ethnic
identification should cause one in his opinions to be more critical of the regime; for Sunnis,
spur one to rally further in its defense. And thus should members of both groups, and here
is Hypothesis 2.2C, exhibit more political engagement and interest as their ethnic identification
increases: for the Shi‘a, this is directed against the state; for Sunnis, over against the Shi‘a—in
order, as al-Ma‘äwdah so helpfully articulated, “to counter probable harm.”
Regarding the prediction of Hypothesis 2.1 that individual material well-being will be a
relatively less significant determinant of political action and opinion in Bahrain as compared to
ethnic group membership, this is not to assert that it will be altogether unimportant or will
not prove a statistically significant predictor of political behavior. The point is merely that
its influence will be one of many competing effects, and of these likely not the greatest in
magnitude; that the role of personal economy in explaining individual political behavior in
rentier regimes has been overemphasized to the exclusion of other likely causes, not that it is
irrelevant. One might ask the question, finally, whether our theory of ethnic-based political
mobilization implies any specific expectations about the effect of economic well-being, more
particularly whether we should expect its effect to be mediated by the ethnic membership of
individuals. Is there a theoretical reason, that is, to think that the effect of personal economy
148
on the political behavior of Sunnis should differ from its effect on that of Shi‘is? In one sense,
perhaps not: our theory, though it explains why Shi‘i citizens will be disproportionately excluded
from public employment and from higher-level government service, does not suggest that
they are completely excluded from such rentier benefits and thus from the clientelism assumed
to alter citizens’ political orientations toward the state. Hence, while the effect of economic
well-being may be presumed to be more systematic among Sunnis inasmuch as they qualify
more universally for employment-related benefits, this difference may or may not be enough
to produce a between-ethnic discrepancy in the effect of household economy itself. This
possibility, however, we shall certainly investigate.
A “Legitimate Aspiration” for All?: Ethnicity and Public Employment in Bahrain
Among the most central and seemingly incontrovertible statements of the foundational rentier
literature is Beblawi’s assertion visited already in Chapter 2 that “Every citizen — if not self-
employed in business and/or working for a private venture — has a legitimate aspiration to
be a government employee; in most cases this aspiration is fulfilled.” After elaborating the
conditions giving rise to the other half of these “most cases,” those cases in which our rentier
citizen is more likely to be rebuffed than have his public employment aspiration fulfilled, at
last we have the opportunity to test this counterargument using individual employment data
from the mass survey of Bahrain. The aim of this analysis is straightforward: to discover
whether ethnic group membership is a significant predictor of individual employment status
among Bahraini citizens, both in terms of one’s sector of work as well as one’s occupational
level. If we find, as anticipated, that one’s chances of being a state employee are reduced
significantly when one is a Shi‘i, and again that Shi‘i ethnic membership is negatively related
to the professional level of one’s pubic-sector position, then we will have evidence that
government employment in Bahrain does not operate neatly in the service of popular political
pacification as rentier theorists would have it, precisely because it disproportionately excludes
those most in need, from the state’s perspective, of pacifying.
At first glance, the data from our Bahrain survey clearly point toward an ethnic-based
discrepancy in public-sector employment. Of the 143 Shi‘i respondents that reported being
employed at the time of surveying, only 55 or about 38.5% worked in the public sector.1 By
1 Respondents were asked exactly this, i.e. “What is your sector of work: public or private?”
149
table 5.1. Individual Sector Employment, by Ethnicity
Sector of Employment Shi‘i Sunni Total
Private
88
51
139 Public 55 52 107
Total
143
103
246
Notes: Pearson’s χ
2 test statistic (with 1 degree of freedom): 3.5221; p = 0.061
contrast, 52 or 50.5% of the total 103 working Sunnis reported being state employees. When
we compute Pearson’s chi-squared statistic measuring the statistical independence of the two
columns, we can comfortably reject the null hypothesis that they are equal, obtaining as we do
an associated p-value of 0.061. That is, if one were to estimate the effect of ethnicity on sector
of employment in a simple probit model, Shi‘i identification would be found to be a negative
predictor of public-sector employment at the 0.061 level of significance. Although a positive
indication, these preliminary results are yet unsatisfactory for two key reasons: first and most
obviously, the association between ethnicity and employment sector may be confounded by
relevant individual-level variables such as gender, education level, and so on. Less obvious
but even more important, moreover, is the fact that our standard probit model, even if were
to include relevant control variables, is not an accurate model of the data-generating process
we are attempting to explain. More specifically, because we only observe values of the sector
variable when a respondent is employed, our sample of the 246 respondents comprising
Table 5.1 above is not a random sample of the Bahraini population but is systematically
truncated to include only the employed Bahraini population. As a result, our apparent between-
group difference in public employment may be a function not of ethnic-group membership
per se but rather of a Sunni-Shi‘i discrepancy in employment in general. That is, there may
be unobserved variables affecting participation in the workforce—women’s participation,
education, etc.—that are also correlated with ethnic group membership and therefore give
us biased estimates of the effect of ethnicity on public sector employment.
For an accurate test of our hypothesis, then, we must adopt an estimation strategy
that reflects both stages of our data-generating process and so avoids the selection bias implied
above: a first that models the probability that a respondent, a random individual from among
the entire Bahraini population, is employed; and a second that models the probability that
this respondent, given that he is employed, is employed in the public sector. Fortunately,
150
we have recourse to Heckman’s (1976) two-step selection model, which carries out exactly this
procedure, designed as it was specifically to correct the sample selection bias problem inherent
in analyses of workforce participation. The Heckman strategy, in short, is a two-equation
structural model that employs one or more identifying variables in the selection equation (i.e.,
the model of workforce participation) to obtain unbiased estimates in the behavioral equation
(i.e., the model of sector employment).2 These identifying variables must be such that they
influence an individual’s chances of being selected (in our case, employed) but do not influence
the outcome of the behavioral model (public-sector employment) except insofar as they do so
via their impact in the selection model. When this condition is satisfied, Heckman’s method
provides unbiased behavioral model estimates no longer influenced by unobserved variables.
In practice, all this is to say that by using a well-known technique we may easily offer
a more robust test of the effect of ethnic group membership on public sector employment in
Bahrain than that of Table 5.1. This (behavioral) model we may specify in the following way:
ελ +⋅+⋅+⋅+⋅+== 43210)1Pr( BBEDUCATIONBFEMALEBETHNICITYBSECTOR ,
where female and education are two control variables that might be presumed related to
public-sector employment in Bahrain, and λ is the inverse Mills ratio calculated from the
selection model (see note 2). Our argument about the role of ethnicity is already known. As
for the controls, their predicted effects we may ascribe to the same cause: to the extent we
2 More generally, the Heckman model offers a correction for sample selection bias by modeling the selection
process using data on those not selected, treating the problem as if it were one of an omitted variable. The selection model is first estimated as a probit model, producing estimated inverse Mills ratio values for each selected case. Using these estimated values, the behavior model is then estimated by generalized least-squares (GLS) regression of Y on the Xs and estimated inverse Mills ratios. This two-step procedure gives us unbiased estimates for B and an estimated ρ, the correlation between the error terms in the behavioral and selection models. More formally, then, the two components of the Heckman model are the behavioral model,
iii BXY ε+= ;
and the selection model,
iii uCWZ +=*, where
1=iZ if 0* >iZ .
This combination, then, gives us the following expression for the expected value of Y, effectively a model of a truncated distribution:
)()0|( *uiiii aBXZYE λρσε+=> ,
where au is the inverse Mills ratio of each observation estimated from the selection model, and λi is the inverse
Mills ratio estimated from the predicted linear (behavioral) model, XB.
151
conceive government employment in the Gulf as an alternative for those unable to find work
in private industry or business, we should expect the state sector to be disproportionately filled
with those less-readily employable (by local standards) elsewhere, including with females and
with those with lower educational qualifications.
What is left, then, is our (selection) model of the determinants of employment proper,
which we may express as the following:
ε+⋅+⋅+⋅+⋅+⋅+== 543210)1Pr( BOLDBMARRIEDBEDUCBFEMBETHNICBWORK ,
where a respondent’s marital status and a dummy for respondents aged 60 and older (a proxy
for retirees) serve as our identifying variables. These indicators, that is, we expect to be
significant determinants of employment status but not of one’s sector of employment per se.
About the predicted effects of these five independent variables perhaps little needs to be said:
the female dummy we should imagine to be a strong, negative predictor of employment;
education, coded on a seven-level scale, a strong, positive predictor; marital status another
very strong, positive predictor of employment (few families would consent to a marriage if the
prospective husband were unemployed; and unmarried women working outside of the home
are relatively rare); and the retiree proxy clearly a strong, negative predictor. The ethnicity
indicator we include again here in the selection model to confirm that its effect on public-
sector employment is not simply a result of its effect on employment more generally. If our
explanation is correct, therefore, the ethnicity variable (coded 0 for Shi‘is, 1 for Sunnis) will
remain a positive predictor of public-sector employment even after controlling its effect on
employment proper. That said, since there is little theoretical reason to believe that ethnic
group membership should be related to wider workforce participation, our prediction in this
case is a non-relationship between ethnicity and working.
The results of estimating this Model 1 are summarized below in Table 5.2. We see that
even after modeling our two-step selection process by Heckman’s method, the influence of
ethnic group membership on public-sector employment in Bahrain remains statistically and
substantively significant. (Compare the biased probit estimates of Model 2.) More specifically,
the marginal effect of ethnicity is a 13.4 percentage point increase in the probability of being
employed in the state sector. In substantive terms, the predicted probability of public-sector
employment (given employment) of two individuals of the same gender and education level
jumps from 38% for a Shi‘i Bahraini to 52% for a Sunni, a relative increase of some 36%. By
152
table 5.2. The Determinants of Public-Sector Employment, estimated two ways
Variables Model 1 Model 2
Heckmana Probit
b
B s
b p >|z| B s
b p >|z|
Behavioral (sector) ethnicity (1 = Sunni) 0.134 0.0685 0.050 0.263 0.168 0.117
female (1 = female) 0.245 0.0973 0.012 0.352 0.192 0.067
education (ascending ordinal) -0.0504 0.0352 0.152 0.00513 0.0671 0.939
married? (dummy) - (omitted) - - (omitted) -
60 or older? (dummy) - (omitted) - - (omitted) - Constant 0.735 0.222 0.001 -0.386 0.351 0.271
Selection (working) ethnicity 0.0430 0.144 0.766 - - -
female -0.868 0.151 0.000 - - -
education 0.355 0.0567 0.000 - - -
married? 0.828 0.152 0.000 - - -
60 or older? -1.460 0.371 0.000 - - - Constant -1.544 0.314 0.000 - - -
Mills Lambda -0.306 0.140 0.029 - - -
Rho -0.731 - - - - - Sigma 0.556 - - - - - Lambda -0.406 0.153 - - - -
a Two-step estimates using the married? and 60 or older? dummies as identifying variables; n = 401, with 238 uncensored b Includes only 245 observations (where working = 1); robust standard errors reported
contrast, every one-point increase on the education-level scale decreases one’s probability of
being employed in the public sector by an estimated 5 percentage points (though the statistical
significance of this estimate is suspect); and being a female increases that likelihood by a
whopping 24.5 percentage points. The predicted probability that a male is employed in the
public sector is just 37%; a female 62%. So in relation to our two control variables, then, the
impact of being a Shi‘i on one’s public-sector employment chances is about one-half that of
153
being male rather than female; or nearly that of a three-point difference on our education scale
(e.g., an elementary school versus college graduate). If perhaps of less substantive importance
than the considerable influence of gender, therefore, yet this observed effect of ethnic group
membership3 on government employment in Bahrain is far from trivial, and lends powerful
empirical support to our critique of the theoretical assumptions underlying the rentier state
framework, in particular to our first hypothesis, Hypothesis 1.1.4
The results of the selection model estimation likewise are on par with our a priori
expectations. We find that female is a strong, negative predictor of workforce participation,
as is our proxy for retirees; whereas being more educated and being married have strong
effects in the opposite direction. Thus we see that our two identification variables—married?
and 60 or older?—are highly-significant determinants of employment, giving us confidence
that we have fulfilled the identification condition of the Heckman model.5 Of more substantive
importance, though, is that the ethnicity variable is shown to be unrelated to employment
per se, confirming that its effect on sector of employment is a direct effect rather than an
indirect one operating via its impact on workforce participation. Finally, we may conclude
from the highly-significant Mills lambda term of –0.306 that the error terms of the selection
and behavioral equations are negatively correlated, which means that the (unobserved) factors
that make employment more likely tend to be negatively associated with the sector variable,
i.e. tend to be negatively associated with public-sector employment. Those qualities other than
3 If the ethnicity variable is excluded from the selection model its behavioral coefficient estimate changes only
slightly to 0.139 (p = 0.034). Neither do things substantively change if only one of the identifying variables is used. 4 In their recent article “Is the Magic Still There? The Use of the Heckman Two-Step Correction for Selection Bias
in Criminology,” Bushway, Johnson, and Slocum (2007) caution users of the Heckman model against, inter alia, employing the two-step estimator versus the maximum-likelihood estimator, which they say is more efficient; and predicting a dichotomous dependent variable in the behavioral model, which they say is explicitly designed as an OLS regression and thus must employ a continuous dependent variable. Since our Model 1 commits both of these supposed blunders, some reassuring words are perhaps required. In short, neither modification makes any substantive difference with regard to our Model 1 here. In the first case, using the maximum-likelihood estimator changes our estimated coefficient on the ethnicity variable from 0.133 to 0.132 and does not alter its z-score. The signs of the control variables education and female are also unchanged and their magnitudes, if reduced somewhat, are relatively unchanged, though the standard error of the former increases a bit to produce an associated p-value of 0.197. As for the authors’ second objection about employing a dichotomous dependent variable in the behavioral model, when we estimate the latter by probit (with the inverse mills ratios from the selection model included manually among the regressors) rather than by OLS via the standard Heckman implementation, again the change makes no difference. Here, in fact, the substantive difference is much less even than that between the two-step and maximum-likelihood estimators: the marginal effect of ethnicity is increased to 0.137, that of female to 0.249, and that of education to –0.0503; and each coefficient has a larger z-statistic than its two-step equivalent. The lambda term remains negative and is significant at the 0.053 level. Far from erroneous, then, compared to these probit results the estimates of our Model 1 are actually conservative. 5 Neither of these is a significant predictor of sector when inserted into the behavioral model.
154
our five selection model variables that make a Bahraini citizen more likely to be employed, in
other words, make him less likely to be employed by the government.
On the one hand, this latter finding would seem to support the common view of public
employment in Bahrain and in the larger rentier Gulf as a sort of sponge with which to sop up
those otherwise relatively less able to find work; yet in that case it only makes even more
glaring the negative influence of Shi‘i ethnic membership on state sector employment, since
according to the selection model results this ethnic membership is altogether unrelated to
one’s likelihood of being employed. In other words, if we make the argument based on our
results that, for example, females and less educated individuals are disproportionately included
in the public sector precisely because they would have a hard time securing work elsewhere,
then we must formulate an entirely different theory to explain why Shi‘is are disproportionately
excluded from the public sector, since we know that after controlling education level, gender,
and so on, they have no harder and no easier a time finding employment than do Sunnis. An
alternative interpretation, accordingly, is that the unobserved factors that make employment
more likely are negatively related to public-sector employment because whereas the private
sector seeks to maximize the productive output of its employees and thus tends to hire on a
meritorious basis, Bahrain’s government sector is more clientelistic than market-driven and
so its hiring process procures fundamentally different employees. Hiring decisions governed
less by candidates’ objective qualifications and more, as Okruhlik says of Saudi Arabia’s state
sector, “by family relations, friendship, [and] religious branch,” the average state employee will
tend to be of a like quality: less desirable by private-sector standards but better connected, better
recommended, and of course better able to produce a “certificate of good history and conduct.”
If such is true it explains why education seems to be unrelated to sector; and indeed it is then
no great mystery why Bahrain’s Shi‘a would be systematically under-represented.
The next step in this analysis is to consider not simply whether one is publically- or
privately-employed but the character and status of that employment. A corollary of the first,
Hypothesis 1.2 predicts that Shi‘i ethnicity will be negatively related to occupational level in
Bahrain. To test this proposition, respondents were asked6 to place themselves on a professional
6 The exact choices were: 1. “employer/manager of establishment with 10 or more employees”; 2. “employer/
manager of establishment with less than 10 employees”; 3. “professional worker (lawyer, accountant, teacher, etc.)”; 4. “supervisory office worker”; 5. “non-manual, non-supervisory office worker”; 6. “foreman, supervisor”; 7. “skilled manual worker”; 8. “semi-skilled manual worker”; 9. “unskilled manual worker”; 10. “farmer, owns farm”; 11. “agricultural worker”; 12. “member of the armed forces, police”; and 13. “housewife.” See Figure 5.3.
155
scale ranging from “an employer/manager of an establishment with 10 or more employees”
to “an agricultural worker.” Appended to this descending scale were two additional options
that eluded easy categorization: members of the armed forces and police; and housewives.
The latter for our purposes we need not consider, but the question of how accurately to
classify military and police personnel is a more difficult one and, as we shall see, one that has
a considerable impact on what we conclude about the effect of ethnic group membership on
occupational level in Bahrain.
This is because, as mentioned briefly already in Chapter 3 (79-80), not a single Shi‘i of
all those interviewed indicated being employed in these services,7 compared to 12 (or 12.1%)
of the 99 working Sunnis who reported their occupations, of whom just one was a female.8
Among employed Sunni males who reported occupational data, then, 11 of 66 (or 16.7%)
said they worked for the military or police, compared to 0 among 117 working Shi‘i males
who reported data. So that, even if we include both sexes, we arrive at an estimate of 1 in
every 8¼ Bahraini Sunnis being employed in the state security apparatus. Moreover, when
we add the data that respondents provided about their spouses, we find that 5 (or 7.2%) of the
69 married Sunnis who reported their spouse’s occupation indicated that s/he worked in the
military or police. Aggregating the two sets of responses, finally, we discover that these 168
observations correspond to just 131 unique Sunni households in which a respondent and/or
a respondent’s spouse was working. This means that of the 131 working Sunni households
in the Bahrain mass survey, a minimum of 17, or 13.0%, are police or military households.
Beyond seeming to vindicate those Bahraini Shi‘a who complain of their exclusion
from the armed services, the fact that we here have a Sunni-exclusive category comprising
some 11.1% of all Sunni respondents is also of more immediate significance, for where we
place it on our scale of occupations will necessarily have a great statistical influence on any
estimated relationship between ethnicity and job level in Bahrain. The question, then, is in
which job category does this group belong? Surely, its present, concluding position below
even farmers and agricultural workers makes little sense. Yet should we deem police and
military personnel “professional workers” of category 3, along with teachers and accountants?
7 One female Shi‘i respondent did report that her spouse worked for the police or armed forces, but since he was
not interviewed one cannot be certain of his ethnicity. The same applies to the analogous discussion of Sunni spouses below. 8 Moreover, two older Sunnis report being members of the armed forces or police but indicate that they are not
currently working, so these two observations are here excluded.
156
1 2 3 4 5 6 7 8 9 11
shi'isunni
Per
cent
age
of R
espo
nden
ts
010
2030
40
“skilled, manual workers” of category 8, along with mechanics? or should the “military/police”
category itself be moved to some other position along the existing scale? Further, given the
heterogeneity of ranks within the military and police, should we assume that all respondents
are lower-level soldiers? commanding officers? or some level in between? Ultimately, such
questions point to the safest course of action, which is to omit the category altogether from
the upcoming quantitative analysis. If we thus lose a bit of statistical leverage on our question
about the relationship between occupational level and ethnicity in Bahrain, at least we shall
avoid making conclusions about it that are unduly influenced by a single category that in any
case seems to be out of place, both physically and theoretically, among the others.
Even with this omission of the military/police sector, however, we face no shortage
of inter-ethnic occupational discrepancies in Bahrain. Consider Figure 5.3, which depicts the
professional categories reported by our employed Bahraini respondents, divided by ethnicity.
Looking at levels 1 through 11, we see that Sunnis are relatively better-represented than Shi‘is
figure 5.3. Occupational Level among Employed Bahrainis, by Ethnicity
157
as “employers/managers of establishments with 10 or more employees” (1); “professional
workers” (3); and “supervisory office workers” (4). By contrast, Shi‘a Bahrainis are relatively
better-represented as “employers/managers of establishments with less than 10 employees” (2);
“non-manual, non-supervisory office workers” (5); “foremen, supervisors” (6); and “skilled
manual workers” (7), and they alone report belonging to the categories of “semi-skilled manual
worker” (8); “unskilled manual worker” (9); and “agricultural worker.” No respondent reported
owning his own farm (10). Certainly, then, we seem to have strong evidence that Bahrain’s
executive, supervisory, and professional classes are relatively better-occupied by the nation’s
Sunnis, and this despite their forming an overall minority of the population.
But as in the case of Hypothesis 1.1, these relative proportions are insufficient to evidence
a general relationship between ethnicity and occupational level. Here we encounter again the
same two problems: the seeming pattern of Figure 5.3 may disappear or change once we add
relevant control variables that also influence one’s occupational level; and our sample is biased
as it includes only those Bahrainis who are employed. Furthermore, the relationship of most
theoretical interest to us is not this between ethnicity and occupational level generally but that
between ethnicity and occupational level among public-sector workers, as per Hypothesis 1.2.
We must therefore employ an estimation strategy similar to but slightly different from that
used already to test our first hypothesis: one that models the selection process inherent in
our occupational data, but one that also can be disaggregated by sector of employment.9 This
two-part model will thus utilize our same selection equation for employment:
ε+⋅+⋅+⋅+⋅+⋅+== 543210)1Pr( BOLDBMARRIEDBEDUCBFEMBETHNICBWORK ,
while the behavioral equation becomes:
ελ +⋅+⋅+⋅+⋅+⋅+= 543210 BBEDUCBFEMBAGEBETHNICBJOBLEVEL if 1=SECTOR .
Compared to Model 1 above, then, the setup is little changed: only the behavioral equation is
estimated now by standard OLS regression given our continuous measure of joblevel. Our
variable of interest, ethnicity, stays the same, as do the control and identifying variables with
9 If the latter condition seems trivial, due to the design of the Heckman selection model, or rather due to its
implementation in common statistical packages, it is not. Using the heckman command in Stata, for instance, if one would attempt to limit the estimation to those cases where sector equals 0, or to where sector equals 1, this would necessarily exclude all cases of unemployed individuals, which simply brings us back to square one and the problem of selection bias. That is to say, to benefit from the Heckman procedure we must apply our sector limitation only to the behavioral model, which we cannot do directly using a standard software implementation. Instead, we must estimate our two equations separately, which is exactly the procedure below.
158
the exception of one additional variable, age, which we might assume to be positively related
to professional level. Also as before, we include the additional regressor λ in the behavioral
equation, which is the inverse Mills ratio for each observation computed from the selection
model (which, again, functions as the control for our sample-selection bias; see supra, note 2).
Substantively-speaking, this estimation procedure corresponds to the following data-
generating process: a random individual from among the adult Bahraini population either is
employed or is not employed, this being determined by the individual’s ethnicity, gender,
education level, marital status, and whether s/he has reached typical retirement age. Having
been thus “selected” for employment, this individual assumes a job that corresponds to a
particular occupational category, and this separate process determining occupational level is
influenced anew by the individual’s ethnicity, age, gender, and level of education. Finally,
this latter process operates differently depending on whether the job in question happens to
be in the private sector or in the public sector; that is, the effects on occupational level of
ethnicity, age, gender, and education are conditioned by sector of employment.
What we should expect to find if our Hypothesis 1.2 were correct, accordingly, is that
the ethnicity variable is a substantively and statistically significant predictor of occupational
level when we restrict the behavioral model to include only public-sector workers. More
specifically, we should expect the relationship between ethnicity and joblevel, which again
is coded in descending fashion from 1 to 11, to be negative: all else being equal, Sunni ethnicity
should cause the expected value of joblevel among Bahrainis to decrease; or, said differently,
Sunni ethnicity should be associated with higher (i.e., closer to 1) professional levels. As for
the control variables, we may offer predictions about their likely effects. Education level and
age one should expect to be strongly associated with higher professional levels for obvious
reasons. The likely effect on occupational level of the female variable, however, is more
ambiguous and depends on our interpretation of its effect witnessed already in Model 1: if we
think that being a female is a strong, positive predictor of public-sector employment because
women are disproportionately excluded from the private sector due to gender discrimination,
then we might expect that the average woman employed in the state sector will be relatively
more qualified than the average male and so will hold a relatively higher occupational position.
On the other hand, if we think that females are relatively better-represented in the public
sector because the public-sector employment process is fundamentally different from that of
159
the private sector, being mediated by unobserved factors such as nepotism, favoritism, and
so on, in this case we have little theoretical guidance to help us predict the influence of gender
on occupational level, which may then be positive, negative, or non-existent.
To gain some empirical leverage on this our primary inquiry, let us first consider the
case of the private sector, which will offer a basis for comparison of our public-sector results.
Here we replicate the estimation procedure described above, simply limiting the behavioral
sample to private-sector rather than public-sector employees (i.e., we specify sector = 0).
We find the results of this Model 3 in Table 5.4 below, which reveals that ethnic membership
has a strong, negative effect on joblevel that is statistically different from zero at a high level
of confidence. As we predicted for the public sector, then, Model 3 indicates that Sunni ethnicity
is associated with higher professional levels, ceteris paribus: among private-sector employees,
being a Sunni rather than a Shi‘i is associated with an estimated 0.582-unit decrease in the
dependent variable joblevel. When we use these estimation results to predict the occupational
category of a Sunni respondent, we find it is approximately 3.9, compared to 4.5 for a Shi‘i.
Sunni ethnicity thus effects a relative improvement in joblevel of a bit more than 15%.
Yet in substantive terms this influence remains somewhat abstract. Let us compare
the effect of ethnicity on occupational level to those of the several control variables. We see,
for example, that the age variable is a highly-significant predictor of occupational level in the
private sector, as one might expect given the time it takes for one to advance professionally.
Its coefficient of –0.0537 tells us that, all else being equal, a one-year increase in a person’s
age corresponds to an estimated 0.0537-unit improvement in job category. If this itself also
means little in substantive terms, yet we see that the effect of ethnicity on joblevel is more
than ten times that of age; that is, Sunni ethnicity corresponds to a 10-year age advantage. So,
to the extent we believe that a ten-year difference in seniority is likely to be of substantive
advantage to one individual over another in terms of professional level in the private sector,
then to a like degree we must acknowledge the significant impact of ethnic membership.
The other Model 3 control variable estimates also warrant mention. Of these perhaps
the least interesting is the education measure, which as one would imagine is an extremely
strong predictor of occupational level, with every one-unit increase in the 7-level education
scale producing, ceteris paribus, a 0.704-unit decrease in joblevel, i.e., upward movement in
professional level. This effect of education level, then, is easily the most important substantive
160
table 5.4. The Determinants of Occupational Level in Bahrain’s Private Sector
Variables Model 3 Model 4
Selectiona OLS
b
B s
b p >|z| B s
b p >|t|
Behavioral (joblevel) ethnicity (1 = Sunni) -0.585 0.319 0.069 -0.461 0.315 0.146
age (ascending continuous) -0.0537 0.0188 0.005 -0.0524 0.0163 0.002
female (1 = female) -0.567 0.478 0.238 -0.270 0.316 0.394
education (ascending ordinal) -0.705 0.204 0.001 -0.858 0.155 0.000
married? (dummy) - (omitted) - - (omitted) -
60 or older? (dummy) - (omitted) - - (omitted) - Constant 9.657 1.727 0.000 10.669 1.018 0.000
Selection (working) ethnicity 0.0357 0.144 0.804 - - -
female -0.887 0.151 0.000 - - -
education 0.359 0.0568 0.000 - - -
married? 0.830 0.151 0.000 - - -
60 or older? -1.464 0.371 0.000 - - - Constant -1.548 0.315 0.000 - - -
Mills Lambda (λ) 0.733 0.963 0.448 - - -
a Manual two-step estimates using the married? and 60 or older? dummies as identifying variables; n = 132 b Includes 135 observations (where sector = 0); robust standard errors reported
impact of all those estimated in Model 3. Next, we notice that the coefficient on the female
control variable, though similar in sign and magnitude to that of ethnicity, is not statistically-
distinguishable from zero at a tolerable level. We must thus conclude that gender, while a
significant predictor of sector of employment itself, nonetheless is unrelated to occupational
level, at least in the private sector. This seems to suggest the absence of any systematic
gender-based discrimination in occupational level among private-sector employees; indeed,
161
if anything the results may be said to support the opposite conclusion, that women, all else
being equal, tend to hold higher-level positions than men.10
Finally we have the lambda term, which one will remember is computed from the
selection model and functions as a control variable that corrects for our sample-selection
bias. As such, its coefficient estimate tells us to what extent our model of occupational
level is affected by unobserved variables operating indirectly via their impact on workforce
participation. In this case, we find that unlike in Model 1, the error terms of our behavioral
and selection equations are unrelated: the unobserved variables that make employment more
likely—i.e., those apart from ethnicity, gender, education, marital status, and being of retirement
age—are not related to occupational level in the private sector at a statistically-significant level.
In other words, estimating occupational level directly, without our selection-model correction,
may not entail sample-selection bias after all. And, indeed, when we estimate our behavioral
model by standard OLS regression (Model 4), the resulting estimates are reasonably close to
those of our more robust Model 3. Yet we notice that the coefficients on the ethnicity and
female variables are now rather lower than before, and that on education higher, meaning
that selection was biasing down the impact of ethnicity and gender on occupational level, and
biasing up the influence of education. In the end, then, our two-stage estimation strategy may
have been overkill, but it has afforded us results in which we may now be more confident.
With these results thus serving as a baseline, let us proceed to consider the more
theoretically-important case of public-sector occupational level in Bahrain. Before continuing
to the estimation results, however, we might first compare the distributions of job level in
the private and public sectors, respectively, to learn how far they appear to follow a similar
pattern. These sector-specific distributions we find below in Figures 5.5 and 5.6. Comparing
the two, we perceive at once that they differ substantially: save for a single Shi‘i respondent,
the categories of 1 and 2 are entirely unoccupied in Figure 5.6, indicating a dearth of director-
level public servants among our sample of respondents who volunteered occupational data;11
further, we see that compared to private-sector employees an overwhelming majority of
10
Of course, this is not to say that there exists no male-female discrepancy at the level of specific job categories or positions (e.g., among job levels 1 and 2 (see note 6), where women are entirely unrepresented, compared to 26 men), but that in the aggregate the average female employed in the non-governmental sector has, ceteris paribus, a higher occupational level than the average male employee. 11
This may be the artifact of a selection effect whereby high-level government employees as a class of respondent were systematically less likely to agree to be interviewed, to answer questions about their employment, to answer these questions honestly, or all the above.
162
1 2 3 4 5 6 7 8 9 11
shi'isunni
Occupational Scale
Per
cent
age
of R
espo
nden
ts
010
2030
4050
1 3 4 5 6 7 11
shi'isunni
Per
cent
age
of R
espo
nden
ts
010
2030
4050
those working in the public sector occupy jobs in either category 3 (“professional workers”)
or 4 (“supervisory office workers”): 62% of Shi‘is and a full 87% of Sunnis. While these two
categories are also modal in the private sector, there we find relatively more variation in job
level among members of both groups, especially among Sunni respondents. In sum, rather
than indicate a common, cross-sectoral distribution of occupational levels, Figures 5.5 and 5.6
seem to reveal two distinct patterns of employment.
This conclusion finds evidence in the results of our regression analysis, summarized
below in Table 5.7. There we see to what extent the determinants of occupational level in
Bahrain’s public sector differ from those in the private sector. An individual’s age, found to be
of such significant import in predicting occupational level in the non-state sector as per our
a priori expectations, we discover is altogether unrelated to one’s professional level in the
government sector, its coefficient having an associated p-value of 0.756. Even with our various
controls in place, therefore, our Model 5 results indicate that the seemingly intuitive assumption
that older individuals will tend to occupy higher-level positions does not obtain in the case of
Bahrain’s public sector. Likewise, we find that a person’s gender, not a statistically-significant
predictor of occupational level in the private sector, in the state sector does indeed play a role,
figure 5.5. Private Sector Occupations figure 5.6. Public Sector Occupations
163
table 5.7. The Determinants of Occupational Level in Bahrain’s Public Sector
Variables Model 5 Model 6
Selectiona OLS
b
B s
b p >|z| B s
b p >|t|
Behavioral (joblevel) ethnicity (1 = Sunni) -0.452 0.293 0.127 -0.431 0.276 0.122
age (ascending continuous) -0.00576 0.0185 0.756 -0.00420 0.0173 0.809
female (1 = female) -0.937 0.443 0.037 -0.764 0.245 0.002
education (ascending ordinal) -0.566 0.193 0.004 -0.651 0.197 0.001
married? (dummy) - (omitted) - - (omitted) -
60 or older? (dummy) - (omitted) - - (omitted) - Constant 7.614 1.257 0.000 8.219 1.555 0.000
Selection (working) ethnicity 0.0357 0.144 0.804 - - -
female -0.887 0.151 0.000 - - -
education 0.359 0.0568 0.000 - - -
married? 0.830 0.151 0.000 - - -
60 or older? -1.464 0.371 0.000 - - - Constant -1.548 0.315 0.000 - - -
Mills Lambda (λ) 0.591 0.699 0.401 - - -
a Manual two-step estimates using the married? and 60 or older? dummies as identifying variables; n = 87 b Includes 90 observations (where sector = 1); robust standard errors reported
and a considerable one at that: its salutary effect on occupational level is to decrease joblevel
by nearly an entire unit; that is, for two public-sector employees differentiated only by gender,
the female is associated with a professional level that is 0.937-units higher (closer to 1) than
that of the male. This coefficient estimate is statistically-distinguishable from zero at a high
degree of confidence. Yet even apart from its magnitude, the fact that gender is at all related
to job level is a significant departure from the results of our private-sector model. As for our
remaining control variables, education and the lambda term, neither differs substantively in
164
its effect on occupational level from that in the private sector: education level is once again a
strong predictor of joblevel, and the non-significance of the lambda term again indicates that
our model of occupational level is not systematically biased by unobserved variables operating
indirectly via their impact on employment per se (hence the similar OLS estimates of Model 6).
We turn finally to our independent variable of most interest, ethnicity. From its
estimated coefficient of –0.452, we see that Sunni ethnicity is again associated with higher
professional levels, all else being equal, and that this effect of ethnic membership is similar to if
somewhat lower than its estimated effect in the private-sector model. More worrying than this
slight drop in magnitude, though, is the statistical confidence of our coefficient on ethnicity,
which has an associated p-value of only 0.127. Yet two things bear mention. First, recall that
for concerns about their proper categorization we have excluded from this public-sector
model all the respondents who indicated that they worked for the military or police. As
each of these 12 respondents was Sunni, the omission of this category is a very influential one
from the standpoint of our estimate of ethnicity. What is more, this exclusion leaves us
with just 87 observations with which to estimate Model 5, compared to 132 for our private-
sector model. By omitting these 12 observations, therefore, we omit a full 12.1% of our
sample, meaning that we have thrown out valuable information that could have helped make
our coefficient and standard error estimates more robust. This latter point leads to a second
consideration: with such a relatively small sample, any outlying observations are rendered
even more influential than they would be otherwise, and thus our anomalous Shi‘i respondent
who reports a public-sector occupational category of 1 (see Figure 5.6) has a disproportionate
impact on our estimate of ethnicity and its standard error, militating against a statistically-
significant relationship between ethnic group membership and public-sector position level.
To better illustrate these concerns, we have in Table 5.8 below the results of several
diagnostic regressions. The first, Model 7, excludes a single observation—our director-level
Shi‘i; while the second and third demonstrate how our results change when we include the 12
military/police respondents: Model 8 codes these responses at the level of category 5, one
above the median category; and Model 9 codes them less conservatively as being equivalent
to category 4. (Note that the outlying observation omitted in Model 7 remains in the latter
two.) In all three of the models, the effect of ethnicity on public-sector occupational level is
increased significantly compared to the Model 5 estimates, and this both in magnitude and in
165
table 5.8. The Determinants of Occupational Level in Bahrain’s Public Sector, Sensitivity Analysis
Variables Model 7 Model 8 Model 9
Omitted Outlier Army = Category 5 Army = Category 4
B sb p >|t| B s
b p >|t| B s
b p >|t|
ethnicity -0.511 0.286 0.078 -0.481 0.293 0.104 -0.716 0.298 0.018
age -0.00139 0.0184 0.940 -0.00841 0.0161 0.604 -0.00686 0.0167 0.681
female -0.999 0.441 0.026 -0.974 0.403 0.018 -0.857 0.411 0.040
education -0.509 0.189 0.009 -0.459 0.140 0.001 -0.352 0.141 0.014 Constant 7.206 1.254 0.000 7.111 0.991 0.000 6.467 0.998 0.000
Lambda (λ) 0.621 0.716 0.389 0.688 0.612 0.264 0.665 0.622 0.287
N 86 99 99
Prob. > F 0.0008 0.0000 0.0016
Notes: selection model results omitted as they are identical to those reported in Tables 5.4 and 5.7; robust standard errors reported
statistical significance. In Model 7, the coefficient estimate on the ethnicity variable increases
in magnitude to –0.511, while its p-value falls to 0.078. Similarly, the Model 8 estimate is a
slightly-smaller –0.481, with an associated p-value of 0.104. In the case of Model 9 the change,
as one would expect, is even more dramatic. In this last estimation the coefficient on the
ethnicity variable balloons to –0.716 and its associated p-value drops to a highly-significant
0.018, demonstrating the decisive impact of the 12 military/police cases, and likewise that of
the choice of how to treat them.12 Finally, we see that none of the changes instituted in these
three diagnostic models affects the estimates of the other independent variables, including of
the lambda term, such as to alter our substantive interpretation of them, confirming that
these 13 cases—the Shi‘i government-sector director and the police/military respondents—
are outliers above all on account of ethnicity and not on the basis of some other variable(s).13
12
Indeed, if these 12 observations are coded as belonging to category 6, the median occupational category, then the ethnicity variable is rendered altogether insignificant, with a coefficient of –0.287 and p-value of 0.334. 13
That said, the magnitude and significance of the coefficient estimates on the female and education controls do drop considerably from Model 8 to Model 9. This is because, in the first place, all but one of the additional cases are males; and, in the second place, because the average education level of these police/military respondents is just 3.67, or somewhere between a primary and secondary school graduate, compared to a mean of 5.18 for the other Sunni state employees in job category 4. The added cases thus dilute the effects of education and gender.
166
What, then, are we to take from this diagnostic testing? and to what extent does it
alter our initial interpretation of Model 5 and the larger question of ethnicity’s impact on
public-sector job level in Bahrain? In short, it would seem that the diagnostic results should
be rather reassuring, both about the robustness of our Model 5 findings as well as about our
earlier decision to omit the police and military cases from our final statistical analysis. On the
first issue, Model 7 has verified in support of our Hypothesis 1.2 the substantive and statistical
importance of the ethnicity marker as a predictor of occupational level among public-sector
employees. If at first we were troubled by the high p-value associated with ethnicity in
Model 5, we understand now that this is an artifact of a single outlying observation that happens
to exert an undue influence over the coefficient and standard error estimates. When this is
omitted we see that, as per our theoretical expectations, Sunni ethnicity is associated with
higher-ranking occupations in the state sector: the predicted job level of a Sunni is about 13%
higher than that of a Shi‘i, all else being equal. As for the matter of the police and military
employees, the extreme volatility introduced by the inclusion of these cases as illustrated by
the Models 8 and 9 results would seem to justify our initial choice to exclude them altogether.
In the end, to preface our results in support of Hypothesis 1.2 with the qualification that these
conclusions are limited to the civilian population only is less damaging than the alternative.
There remains, finally, a lingering theoretical issue that, while not tied directly to the
two hypotheses investigated in this section, has arisen naturally over the course of our analysis
here. This is the more general question of how properly to conceive public-sector employment
in Bahrain, or in the class of Gulf rentier states, independent of the effect on this of ethnicity.
Do the preceding results suggest any coherent conclusions about the bases of public-sector
vis-à-vis private-sector employment? Can they tell us, that is, what sort of citizen tends to be
employed by Gulf governments? and why? So far we have evaluated our empirical results in
light of two competing interpretations: a first that views state employment as a sanctuary for
those less suited for the private-economy workforce; and a second that sees it not as a labor
market correction but as a political tool in its own right. If the latter would seem plausible
enough, recall that the former is much closer to the notion articulated by rentier theorists, in
particular to Beblawi’s parenthetical qualification that “Every citizen — if not self-employed in
business and/or working for a private venture — has a legitimate aspiration to be a government
employee” (emphasis added). Yet, as pointed out already, we have several pieces of evidence
seeming to run counter to this standard rentier understanding.
167
In the first place, apart from a person’s ethnicity only gender seems to be a statistically-
significant determinant of public-sector employment among Bahrainis, whereas if Beblawi’s
interpretation were correct we would also expect education level to be a strong, negative
predictor of state employment.14 Moreover, since we found that ethnicity has no independent
effect on employability per se, it is difficult to square the strong, negative relationship between
Shi‘i identification and public-sector employment with the idea that the latter serves mainly
as a fallback for those otherwise unable to secure work. Looking now to our Hypothesis 1.2
testing, we find similarly that the results are inconsistent with the idea that government jobs
tend to be provided to those less able to find employment elsewhere. For one thing, gender
is shown to be unrelated to a person’s occupational level in the private sector while female
employees in the public sector hold systematically-higher positions than do males of identical
education, age, ethnicity, and so on. Even more strangely, the notion of workplace seniority
wherein, ceteris paribus, older employees will tend to occupy higher-ranking jobs insofar as
they are likely to be better-endowed with professional experience and the intangible skills it
implies, obtains only in the private sector, whereas in the public sector an individual’s age is
entirely unrelated to occupational level. Yet at the same time, education level is the single
most important predictor of occupation in both sectors, and Shi‘i ethnicity likewise serves as
a professional hindrance in each; so why is the public sector in Bahrain responsive to these
two “market” forces but not to the influences of age or gender?
Though our analysis above can provide no direct answer, these facts do suggest that
the process by which a Bahraini comes to gain employment in a given position operates in a
fundamentally different way depending on whether that position is in the public or private
sector, a conclusion also supported by our Model 1 finding that the unobserved factors that
make employment more likely are negatively associated with public-sector employment. As
discussed already, one alternative interpretation based on these observations, therefore, is
that whereas employment and occupational advancement within the private sector tends to
proceed more according to one’s objective qualifications (notwithstanding the impediment
of ethnicity), that in the public sector operates along qualitatively-different lines: favoritism,
14
If one would wonder whether Beblawi’s conception of government employment and employees is indeed as suggested here, he need only consider the sentences that follow the quotation above, which read: “Though utterly free enterprise oriented, the number of government employees in the oil states is only matched by socialist-oriented states. Civil servant productivity is, understandably, not very high and they usually see their principal duty as being available in their offices during working hours” (91).
168
personal and family relationships, and notions of national loyalty and one’s being “deserving”
of state employment. Whether or not one is satisfied with this alternative explanation, it is
clear in any case that the original understanding of Beblawi and other rentier theorists, the
notion that in allocative regimes the public sector serves as a sort of parallel job market that,
if less demanding of its aspirants, remains fundamentally economic-based (or “utterly free
enterprise oriented”; cf. supra, note 14) and politically-agnostic—it is clear that this conception cannot
tell the whole story. Instead, here as elsewhere we find evidence that the ostensive material
underpinning of the rentier-based economy, the cold economic bargain said to exist between
Gulf rulers and their clients-cum-citizens, is a rather more pragmatic partnership, colored and
adulterated by other, competing considerations. In Bahrain at least, the smooth translation
of rentier wealth into tangible material benefits for all citizens seems not to operate so smoothly
after all. How this fact serves to alter the final step in the supposed rentier equation—the
translation of material gains for citizens into political dividends for rulers—we shall now see.
The Determinants of Individual Political Opinion and Action in Bahrain
The aim of this second section is to provide evidence from the Bahraini survey in support of
Hypotheses 2.1 and 2.2, which together embody our critique of the notion that untaxed rentier
citizens make politically-passive rentier citizens. The thrust of this critique is, again, that an
argument positing a relationship between one’s tax burden and one’s political behavior is not a
theory of rentier politics but rather a single empirical prediction, and as such one that can tell
us little about the overall political orientations of citizens in rent-based states insofar as there
likely exist other, non-material determinants of political action and opinion. That one did not
vote in order to protest a new tax hike, in other words, does not mean that one did not vote.
So too with our politically-active Bahrainis, for whom, if one is convinced by the account of
Chapter 3, at stake is something perceived to be larger than individual economic prosperity.
To keep our analysis as parsimonious as possible, we will test the several predictions
that comprise Hypotheses 2.1 and 2.2 simultaneously within a standard, comprehensive model,
which will be employed to investigate first the individual-level determinants of political opinion
among our Bahraini respondents followed by those of political action. As Hypotheses 2.1A and
2.1B simply imply a significant between-group difference in the dependent variable, we will
test these with the inclusion of a dichotomous ethnicity indicator such as was used in the
previous section’s analysis of employment. Hypotheses 2.2A through 2.2
C, on the other hand,
169
make a relatively more complicated argument: that political opinion and action in Bahrain is
influenced not only by ethnic group membership per se but also by the strength of one’s ethno-
religious identity as measured by individual religiosity; and that, moreover, the effect of this
religiosity itself depends on whether one is a Sunni or a Shi‘i. What Hypotheses 2.2A-C
imply,
that is, is that the political orientations of Bahrainis are determined in part by an interaction
between the effects of two variables: one’s ethnic identity per se and the personal salience of
that identity as measured by one’s religiosity,15 a claim we may evaluate by including in our
model a multiplicative interaction term ethnicity × religiosity.
As for the measure of respondent religiosity itself, we have several alternatives. One
commonly-used measure is the frequency with which one performs religious rituals such as
prayer, reading of the Qur’än, or mosque attendance. The difficulty with such measures,
however, is that due to the practical differences between Shi‘i and Sunni Islam, the substantive
meaning of, say, “mosque attendance” is likely to differ across the two groups. For a Bahraini
Shi‘i, for example, this may be taken to include religious services at a ma’tam or for a religious
commemoration. Likewise, Shi‘is, who in Bahrain as elsewhere seek edification in “Husaynï”
literature ( حسينيةبكت or األدب احلسيين), may report reading the Qur’än less than Sunnis of equal
religiosity, rendering that measure inconsistent. Moreover, all but overtly secular respondents
will probably tend to exaggerate the frequency of their observance lest they appear irreligious,
an incentive operating upon Sunni and Shi‘i alike. For the same reason one must be careful
too in employing the other standard instrument used in survey research to gauge religiosity,
which is to ask the respondent directly: “In general, would you describe yourself as a person
who is religious ( نمتدي ) or non-religious ( نغري متدي )?” In the first place, once again apart from
deliberately-secular individuals, a typical Arab Muslim of any religiosity is unlikely to wish to
describe himself in this fashion as “non-religious” and so will tend to default to “religious.”
In the second, the most commonly-employed word “religious” (نمتدي) here carries overtones in
Arabic of someone who is not simply religiously-observant but unusually devout and perhaps
even fanatical in his faith. It is no wonder, then, that of the 389 Bahrainis who responded to
15
One might wonder why we would resort in gauging the strength of a Bahraini’s ethnic identity vis-à-vis the rival group to a general measure of religiosity as opposed to, say, by asking a respondent directly about his views of, trust in, or relationship with members of the other ethnic group. Assuming one could have elicited truthful responses (and this is a questionable assumption), such indeed might have been preferable. In fact, this has been the procedure elsewhere, for example in the 2004 and 2006 surveys of Iraq to be dealt with in the next chapter. Yet how the Iraqi field interviewers managed to avoid being sent away by or irreparably offending interviewees upon asking such pointed questions remains a mystery to me (although their avoidance of 4 of Iraq’s 18 governorates may be part of the answer), for I am sure that we would not have been so fortunate in Bahrain.
170
this question, 57 (or 15%) were unsatisfied with both choices and so asked the interviewer to
record an alternative answer: “معتدل,” or “moderate”; and one assumes that many others who
were less concerned for the accuracy of their response simply picked one or the other option.
An alternative method of measuring religiosity, then, is to infer it indirectly from one’s
response to another, less transparent question. If the resulting indicator enjoys perhaps less
outward validity than one based on a direct inquiry, still it is less affected by incentives to
misrepresent oneself and, in reference to the first class of indicators discussed above, is also
consistent across the two religious subgroups. The religiosity measure we will adopt for our
analysis is of this final type, constructed on the basis of the question: “Which of the following
factors [would] constitute the most serious impediment to your acceptance of the marriage of
your son, daughter, sister, or brother?”:16 “lack of (عدم) prayer [on the part of the betrothed],
lack of fasting, the social status of the [betrothed’s] family, poverty, lack of education, lack of
employment, or other.” Our religiosity variable is coded 1, then, when a respondent identifies
either or both of the first two factors as being the most important or adds an “other” category
response that invokes religion. Among the latter instance are included, for example, responses
that the most important attribute of the betrothed is reputable “ethics” or “morals” (األخالق), “religiousness” (نالتدي), or simply “belief in God.” Several individuals even state explicitly that
the most important thing is that s/he be a Sunni/Shi‘i. That the “other” category would in
this way be filled almost entirely with religious stipulations, including with ethno-religious
stipulations, speaks well for the validity of this measure as a proxy for an individual’s level of
religiosity.17 As a diagnostic check, finally, we may substitute for this religiosity variable the
more conventional “Are you religious?” indicator to confirm the robustness of our results.
The other independent variable of interest here is a respondent’s material well-being,
whose relative impact on political opinion and action in Bahrain is predicted in Hypothesis 2.1
to be less than or equal to that of ethnic group membership in substantive and statistical terms.
Later we asked (cf. supra, 147-148) whether there was theoretical cause to expect that the influence
16
The exact wording of the question is:
“ أخوك؟/أختك/ابنتك/أي من العوامل التالية يشكّل العائق األكثر أمهية أمام موافقتك علی زواج ابنك ” 17
By this measure, 268 of 401 (or 69% of) total respondents are coded as “religious,” and 133 as “not religious.” Compare this to the 221 of 389 (or 57% of) individuals who self-identified as “religious” when asked directly. (If we include those who replied “moderate” this rises to 71%). Yet the correlation between the two measures is a relatively-low 0.263 (or 0.295 if the “moderates” are excluded), meaning that on the whole the sort of person who self-identifies as “religious” when asked directly is not the same sort of person who exhibits concern for the religious in the case of what is in the Arab world perhaps life’s most significant practical matter: marriage.
171
of this economy indicator should vary across our two ethnic groups. While we ultimately
replied that there probably is not, we noted that the disproportionate exclusion of Shi‘is from the
state sector (and thus from the rentier benefits of public employment) may serve to dampen the
link between economic well-being and political orientations among Bahraini Shi‘a as compared
to among Sunnis, who as a group have a more reasonable expectation of receiving government
benefits in return for (outward) political allegiance. To evaluate both of these predictions
regarding the effect of household economy,18 therefore, we may simply repeat the procedure
above to include in our model of political behavior a standalone economy variable along with
the interactive term economy × ethnicity. And as for the specific claim of Hypothesis 2.1, we
may easily assess it by comparing the relative influence of ethnicity to that of economy.
Concerning control variables, we will include of course the standard indicators of age,
female, and education that have been utilized already in our study of employment. The
first we may expect to be negatively related to political action (i.e., to be associated with less
frequent political action) and positively related to political opinion (i.e., to be associated with
more favorable or pro-government political opinions) inasmuch as we may expect that people
will tend to become more risk-averse as they age and acquire things—families, homes—the
enjoyment of which could be jeopardized by political dissent or activism. Women we should
also suppose to be less likely to engage in political action given the persistence of traditional
gender roles in Gulf society; yet we have no a priori reason to assume that Bahraini women
should be more subdued in their political opinions than men, and, indeed, my experience in
Bahrain would sooner suggest the opposite. As for the expected effect of education, lastly,
little needs to be said: higher levels of respondent education should be associated with more
frequent political action and more negative political opinion.
Beyond these basic demographic measures, moreover, we will include in our model
of political behavior in Bahrain a final control variable alluded to in our discussion of survey
procedure in Chapter 4: a dummy indicating whether a respondent happened to be interviewed
by a fieldworker of the other ethnic group. This control, coded dichotomously as diffethnic,
18
Note that, as its name implies, our indicator here is not a measure of wealth per se but one that captures the overall economic well-being of a respondent. This is in order to remain faithful to the rentier causal story, whose predictions about individual political behavior are based not on finite measures of citizens’ income or prosperity but according to their overall economic satisfaction. Thus we might just as easily have named our economy variable satisfaction. The exact instrument from which this indicator is taken reads as follows:
“How would you rate your family’s economic situation right now?: : القتصادي ألسرتك اآلن؟كيف تقيم الوضع ا ”
1. Very Good, 2. Good, 3. Bad, 4. Very Bad.” “ سيئ جداً. ٤ ،سيئ. ٣ ،جيد. ٢ ،جيد جداً. ١
172
will allow us to test the suggestion of Chapter 4 that respondents’ answers, in particular their
answers to sensitive socio-political questions, should be significantly influenced by the ethnicity
of their interviewer; and, more generally, it offers a straightforward metric by which to gauge
the extent of ethnic mistrust in Bahrain. Moreover, if we interact this diffethnic variable with
ethnicity, we may also learn something about the nature of that mistrust: specifically, if we
find that both Sunni and Shi‘i respondents of inter-ethnic interviews tend to give politically-
safer answers as compared to same-ethnicity interviewees, if they tend alike to report having
engaged in less political action and holding more pro-government opinions (or, said differently
still, if the interaction term diffethnic × ethnicity is not a significant predictor of responses),
then we may conclude that this apparent mistrust of the ethnic other in fact is something
more akin to mistrust of the survey project itself (that is, uncertainty whether it is not perhaps a
government effort to identify dissenters), with the unfamiliar ethnicity of the field interviewer
serving only to increase a respondent’s overall discomfort and suspicion.
But if, on the other hand, we find that the effect of inter-ethnic interviewing varies
across ethnic groups, if we find that Bahraini respondents do not uniformly moderate their
answers to assume more pro-government positions but instead alter them to be more in line
with the presumed, ethnically-ascribed political views of their interviewers, then we have
evidence of an entirely different dynamic. In this latter case, we must conclude that inter-
ethnic interviewees are indeed influenced by the ethnicity of the interviewer per se and not
by the interviewer’s foreign ethnicity as a proxy for the overall dubiousness of the interview.
Though in both instances the respondent tends to offer a “safer” answer as compared to his
unaffected response, here his answer is not safer vis-à-vis the state but vis-à-vis his interviewer,
to whom he ascribes specific, ethnically-defined political views that he understands differ
exactly for this reason from his own.19 Precisely when and why an individual will tend to
misrepresent his own views and actions to avoid confrontation with another who is known
to differ is a question for psychology. The important upshot for our purposes is that, should
this be found the case, we will have evidence of an ethnic division in Bahrain that not only
19
There remains the alternative, of course, that respondents of one or both ethnic groups are unaffected by this inter-ethnic interviewing, in which case Bahraini Sunnis, Shi‘is, or both tend to remain consistently either honest or dishonest in their responses irrespective of interviewer. But note that so long as the diffethnic variable is significant, we should always expect it to be related to more positive political opinion and to less political action (answers that may be “safer” vis-à-vis the government or vis-à-vis the other-ethnic interviewer), meaning that it is the sign and significance of the diffethnic × ethnicity term that must decide between our competing interpretations. See Table 5.9 below.
173
table 5.9. The Determinants of Political Opinion and Action among Bahrainis, Predictions
Variables Opinion (improving) Action (increasing)
Relationship Significant?a Relationship Significant?
a
ethnicity (1 = Sunni) + Yes – Yes
diffethnic (1 = Yes) + Yes – Yes
diffethnic × ethnicity ? Yes ? Yes
age + Yes – Yes
female None No – Yes
education – Yes + Yes
economy (descending) – ? + ?
economy × ethnicity – ? + ?
religiosity (1 = religious) – Yes + Yes
religiosity × ethnicity + Yes None No
? indicates ambiguous theoretical expectations a
Note that the marginal effects and the statistical significance of those effects for the interactive term components—i.e., all variables but the controls age, female, and education—will not be given simply by the reported coefficients and standard
errors of our model output as per usual. Rather, since our theory states that the effects of these variables on individuals’ responses with vary depending on the values of some other variable(s) (namely, ethnicity), we must estimate these marginal effects and standard errors manually over a substantively-meaningful range of our modifying variable(s). See the instructive 2006
Political Analysis paper “Understanding Interaction Models: Improving Empirical Analyses” by Brambor, Clark, and Golder.
influences individual political opinion and action but even transcends the political sphere to
alter basic social interaction between members of the two ethnic groups.
Taken together, the above-described variables give us the following standard model
of individual political behavior in Bahrain:
+⋅+⋅+⋅×+⋅+⋅+= 543210 BFEMBAGEBETHDIFFETHBDIFFETHBETHNICBRESPONSE
ε+⋅×+⋅+⋅×+⋅+⋅ 109876 BETHRELIGBRELIGBETHECONBECONBEDUC ,
where response is a given dependent variable of interest. For variables coded continuous, the
model is estimated by ordinary least-squares regression; for those that are ordinal, by ordered
probit; and for those that are dichotomous, by probit. Before commencing our analysis, we
174
may review our theoretical predictions about the effects of our ten independent variables on
the political opinion and action of ordinary Bahrainis. These are summarized above in
Table 5.9 and should require no further elucidation. Note finally that in addition to our
combined standard model here we may wish to carry out ethnically-segregated estimations
(i.e., one model for Sunnis and one for Shi‘is) in order to obtain separate, ethnic group-specific
estimates for all the independent variables. In that case ethnicity and the three interaction
terms will drop out of the equation due to collinearity, leaving just six independent variables.
Tempered by Non-Taxation?: The Sources of Bahraini Political Opinion
To clarify our generic term employed thus far, we may say that the “political opinions” we shall
investigate here are of a particular type. These are not, in the first place, general attitudes
toward political systems (“Is democracy better than the alternative?”), political tolerance
(“Do women make good leaders?”), the place of religion in politics (“Should clerics be able
to influence the way people vote?”), and so on. The indicators we shall use are not, in other
words, hypothetical, regime-agnostic questions about political values or orientations. While
the answers to these questions may be interesting for other reasons, and while such questions
were duly asked of the Bahraini respondents as were all those of the Arab Barometer survey
instrument, nonetheless the theoretical argument propelling the present inquiry forward has
little to say about this category of opinion. Instead, the rentier theory we wish to evaluate
makes the specific claim that in allocative economies citizens will tend to be more passive and
quiescent politically in proportion to the material benefits that accrue from the rentier state, a
prediction about citizen-state interaction rather than about citizens’ innate political natures or
ideals.20 The indicators we shall select, then, are meant to represent popular attitude toward
the regime itself, toward the basic legitimacy and desirability of the prevailing political order in
Bahrain. As such they will measure the following opinions of citizens: (1) the quality of the
nation’s overall “political situation” (“الوضع السياسي”); (2) the government’s overall positive or
negative influence on a respondent’s daily life; (3) the legitimacy of the 2006 parliamentary
20
Of course one might also make the latter argument, as indeed the early rentier literature tends to do in its talk of a deleterious “rentier mentality” affecting the citizens of rent-based regimes. Defined generally as “a break in the work-reward causation” (Luciani 1990, 88)—that is, an aversion to (physical) work coupled with a sense of material entitlement—this complex of citizen values is posited as one of the very defining characteristics of an allocative state. It is an interesting question whether such a change in culture, could it occur at all, could have occurred already in the timeframe of but a few generations since the foundation of the modern rentier states. Yet it is a question that demands an altogether separate inquiry.
175
Very Good Good Bad Very Bad Not Sure Refuse
shi'isunni
Per
cent
age
of R
espo
nden
ts
010
2030
40
elections; (4) the degree of respect for human rights in Bahrain; (5) one’s level of trust in
important state institutions; (6) one’s overall satisfaction with “government performance”
.and (7) one’s level of national pride ;(”أداء احلكومة“)
The first variable, political situation, derives from what is perhaps the most direct
and general political question of the entire survey, appearing very early in the interview: “In
general, how would you rate the present political situation in the country?”21 The response
categories descend in the standard manner from “very good,” “good,” “bad,” to “very bad.”
As expected, we observe in Figure 5.10 above a drastic between-ethnic difference in response:
whereas a majority (a combined 56%) of Sunni respondents report that Bahrain’s present
political situation is “good” or “very good,” Shi‘is are tilted even more in the opposite direction,
with a full 71% of respondents describing the political situation as “bad” or “very bad.” Indeed,
21
Respondents are asked in Arabic: “بشكل عام، كيف تقيم الوضع السياسي احلايل يف البالد؟”
figure 5.10. Opinion of the Overall Political Situation in Bahrain, by Ethnicity
176
some Shi‘a even preferred in lieu of “very bad” to give still more emphatic responses such as
“ or (”ruin [in]“) ”دمار“ البحرينما يف سياسة يف ”—literally, “there is no politics in Bahrain,” implying
a total domination of political decision-making by the Äl Khalïfa. Finally, it is clear from the
relatively high rate of “I’m not sure” responses that many individuals, especially Sunnis, were
wary of answering this question altogether on account of its overt politicality.22 Overall,
though, the utter inversion of the red and black bars for the valid responses seems to offer an
excellent visual summary of Bahrain’s ethnic divide surrounding the political status quo.
Below in Table 5.11 we find the results of the more complete analysis employing our
standard model of political opinion in Bahrain. In addition to this ordered probit estimation
of Model 1 we have two diagnostic models to verify the robustness of these results: a first uses
the alternative religiosity measure described above—the more common “religious” or “not
religious” indicator, including the “moderate” responses in between—and a second estimates
the standard model by ordinary least-squares. The main benefit of the latter is that, even if it
makes less efficient use of our categorical dependent variable political situation, giving us
perhaps more conservative coefficient and standard error estimates, still it allows for a more
straightforward interpretation of the marginal effects of our independent variables of interest.
We may begin by observing that all three models appear to offer substantively-equivalent
results: the signs of all the coefficient estimates remain the same, and their significance too is
little changed from Model 1 to Model 3. Yet recall (cf. Table 5.9, note a) that only the effects
of our three control variables—age, female, and education—are directly interpretable from
the raw model output of Table 5.11. This is, once again, a necessary result of our conditional
hypotheses tested here by the several interaction terms: the latter imply by definition that
the effect of one variable (say, religiosity) on individual political opinion is itself dependent
upon another condition (namely, one’s ethnicity). It makes little sense, then, to speak of an
unconditional marginal effect of religiosity on political situation, which is the coefficient
0.327 reported in Model 1, when our theory tells us that this effect is not unconditional but will
22
Ideally, one would prefer to make use of these “I’m not sure” and “refuse to answer” responses rather than just exclude them from model estimations, since these markers of apprehensive respondents may provide helpful measures of individuals’ propensities to give honest answers. One might, for example, construct a variable based on the frequency of an individual’s “I don’t know” or “refuse to answer” responses to selected sensitive questions. The difficulty, of course, apart from distinguishing a reluctance to answer from genuine ignorance or indecision, is that for those questions where such a variable would be most useful as a control—i.e., for the most sensitive questions—the most apprehensive individuals are disproportionately likely not to appear in the sample at all, since they will have responded “I don’t know” or refused to answer with a relatively high probability. Thus our measure of apprehension may be hampered by a selection effect of this sort. While we do not include such a “refuse” variable in our standard model, then, we do employ one later as a diagnostic control. Cf. infra, 270 ff.
177
table 5.11. Bahrainis’ Opinion of the Country’s Overall Political Situation, estimated three ways
Variables Model 1 Model 2 Model 3
Standard Alt. Religiosity Measure OLS Regression
B sb p >|z| B s
b p >|z| B s
b p >|t|
ethnicity -1.009 0.474 0.033 -1.050 0.474 0.027 -0.750 0.318 0.019
diffethnic -0.783 0.180 0.000 -0.764 0.181 0.000 -0.519 0.120 0.000
diffeth × eth 0.830 0.330 0.012 0.985 0.317 0.002 0.547 0.229 0.017
age -0.0111 0.00466 0.017 -0.0133 0.00453 0.003 -0.00739 0.00316 0.020
female 0.0740 0.123 0.547 0.0741 0.119 0.533 0.0451 0.0816 0.581
education 0.0892 0.0476 0.061 0.107 0.0466 0.022 0.0614 0.0328 0.062
economy 0.396 0.143 0.006 0.409 0.146 0.005 0.242 0.0864 0.005
econ × eth -0.0524 0.199 0.793 0.0366 0.206 0.859 -0.00525 0.130 0.968
religiosity 0.327 0.175 0.061 0.191 0.0850 0.025 0.213 0.117 0.069
relig × eth -0.498 0.266 0.061 -0.366 0.142 0.010 -0.335 0.182 0.067 Constant - n/a - - n/a - 2.497 0.330 0.000
N 357 347 357
Prob. > F (χ2) 0.0000 0.0000 0.0000
(Pseudo) R2 0.1624 0.1620 0.3466
Note: Robust standard errors reported for all models
depend rather on a person’s ethnic affiliation. Therefore, to interpret properly the impact on
political opinion of our four explanatory variables of interest—ethnicity, diffethnic, economy,
and religiosity—we will offer estimates of the magnitude and statistical significance of their
marginal effects over a substantively-meaningful range of the appropriate modifying variable
as suggested by Brambor, Clark, and Golder (2006).23 Except when evaluating the effect of
ethnic group membership itself, this modifying variable will be the ethnicity indicator, whose
“substantively-meaningful range” is of course 0 (Shi‘i) and 1 (Sunni).
23
More than just for their instructive paper, I must thank these authors for their indispensible Stata code (available at http://homepages.nyu.edu/~mrg217/interaction.html) that makes presenting these effects much easier.
178
We will begin by analyzing the marginal effect of ethnicity on respondents’ opinion
of the overall political situation in Bahrain, which should reflect the dramatic between-group
difference revealed already in Figure 5.10 above. It is tempting simply to point to the large and
statistically-significant coefficient on the standalone ethnicity indicator in Table 5.11 and from
this conclude that ethnic group membership is indeed an important predictor of Bahrainis’
views in this case. Recall, though, that according to our theory the effect of ethnicity should
depend on the religiosity of an individual, with more religious Shi‘is tending to give less
favorable and more religious Sunnis tending to give more favorable opinions. Similarly, the
effect of ethnicity also operates in our model via its impact on diffethnic and economy (i.e.,
via the two interaction terms diffethnic × ethnicity and economy × ethnicity).24 In sum,
the effect of ethnicity will vary across individual respondents depending on their religiosity,
economic satisfaction, and whether they were interviewed by someone of the other ethnicity.
For the purposes here of testing Hypotheses 2.1 and 2.2, however, we are less interested
theoretically in ethnicity’s influence on political opinion as one’s economy changes or in the
case that one is part of an inter-ethnic interview. Instead, we are interested in the opposite
relationships: the effect on political opinion of economy and inter-ethnic interviewing as
respondent ethnicity changes (hypothetically, that is) from Shi‘i to Sunni. In computing the
marginal effect of ethnicity, therefore, our modifying variable of interest is religiosity, and
will remain such throughout the section. So as not to exclude the impact of ethnicity via
economy and diffethnic, though, we fix the latter two variables at their mean values, which
are approximately 2.08 and 0.232, respectively. That is, we will compute the effect of ethnicity
for an individual of “average” economy and “average” diffethnic (though the latter is not
substantively meaningful).25 All this is to say the following: we wish to discover the extent of
ethnicity’s impact on political situation: does this influence obtain only for individuals of
high religiosity, low religiosity, or both?; and, in the final case, is its magnitude indeed greater
among individuals of higher religiosity as per our Hypotheses 2.2A and 2.2
B?
24
More exactly, the marginal effect of ethnicity on political situation is the first derivative of ethnicity with respect to the dependent variable, political situation. In our model, then, this gives us:
10831)(
)(BYRELIGIOSITBECONOMYBDIFFETHNICB
ETHNICITY
SITUATION ⋅+⋅+⋅+=∂∂
. 25
One might ask why we do not choose to set diffethnic at 0—i.e., why we do not choose to compute the effect of ethnicity when one’s responses are entirely unaffected by inter-ethnic interviewing. For one thing, we do not yet know the precise impact of the latter. For another, we may always adjust this value later for illustrative purposes. In any case, its current value of 0.232 serves only to make more conservative our estimates of the marginal effect of ethnicity, which is no great harm.
179
-1.5
-1-.
50
.5
Mar
gina
l Effe
ct o
f Eth
nici
ty
Not Religious Religious
Religiosity of Respondent
Marginal Effect of Ethnicity95% Confidence Interval
Dependent Variable: Overall Political Situation
Marginal Effect of X on Y As Z Changes
model
We find in Figure 5.12 above a plot of the marginal effect of (Sunni) ethnicity on a
Bahraini’s opinion of the overall political situation in the country.26 The y-axis represents the
magnitude of this effect and the x-axis represents the (0 to 1) range of individual religiosity.
The estimate of the effect of ethnicity is statistically-significant at the 0.05 p-level whenever
both the upper and lower bounds of the 95% confidence are either below or above zero. In
this case, we see that its effect is significant across the entire range of religiosity—that is, it
is a statistically-significant predictor of political situation irrespective of whether one is a
religious individual according to our measure. More specifically, the marginal effect of one’s
being a Sunni rather than a Shi‘i ranges from about –0.63 among non-religious respondents
to –0.97 among religious respondents. In terms of our dependent variable, this corresponds
to a change of around two-thirds of one category in the first case and to an entire category in
the second. Sunni ethnicity, in other words, is associated among religious individuals with a
change from, say, “bad” to “good” or “good” to “very good,” when asked about the country’s
26
It should be clear by now that the marginal effects reported here correspond to the OLS-estimated model (Model 3 in Table 5.11) rather than the ordered probit model. This is to allow a more intuitive substantive interpretation and, at worst, simply renders our marginal effect estimates more conservative.
figure 5.12. Marginal Effect of Ethnicity on Opinion of Overall Political Situation
180
1:
Ver
y G
ood
2: G
ood
3: B
ad4:
Ver
y B
ad
Opi
nion
of O
vera
ll P
oliti
cal S
ituat
ion
Non-Religious Religious
Religiosity of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
political situation as compared to a Shi‘i respondent of identical economic satisfaction, age,
gender, education, and so on.
Figure 5.13 above may provide a more intuitive depiction of our findings by showing
predicted values of our response variable political situation for Sunni and Shi‘i respondents
according to their religiosity and after controlling for all other explanatory variables.27 The
predicted response of a Shi‘i respondent is shown to be just shy of category 3 (“bad”) among
non-religious individuals and to cross further into category 3 towards category 4 (“very bad”)
as religiosity increases. Conversely, the predicted response of a Sunni Bahraini hovers around
category 2 (“good”) for non-religious individuals, moving closer toward category 1 (“very
good”) as religiosity increases. Finally, we may repeat this exercise using our alternative
measure of religiosity—“Would you describe yourself as religious or not religious?”—in order
to verify that the pattern above is not an artifact of the way we measured our modifying
variable. Figure 5.14 below gives these new predicted values, which if anything reveal an even
27
Note that in this and the other predicted values plots of this section, all control variables are standardized.
figure 5.13. Predicted Values of Response Variable, by Ethnicity and Religiosity
181
1:
Ver
y G
ood
2: G
ood
3: B
ad4:
Ver
y B
ad
Opi
nion
of O
vera
ll P
oliti
cal S
ituat
ion
Non-Religious Moderate Religious
Religiosity of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
greater divergence between Sunni and Shi‘i respondents as they increase in religiosity: here
we see that the line describing the predicted response of a Shi‘i Bahraini is shifted upward as
compared to that in Figure 5.13, while the line describing Sunnis has a negative slope that is
noticeably greater in magnitude than that of Figure 5.13, even crossing below the category 2
threshold as it approaches the far end of the religiosity scale. In other words, our substantive
interpretation remains the same regardless of the modifying variable religiosity we employ;
our conclusion—that Shi‘i ethnicity in Bahrain is associated with less favorable opinions of
the country’s overall political situation, and that this between-group divergence in opinion
increases with respondents’ religiosity—is robust across both measures.
Having learned of the impact on political situation of ethnic group membership,
we now proceed to analyze the effects of our other independent variables of interest, namely
diffethnic, religiosity, and economy. We may begin with the first. Depicted in Figure 5.15
below is the marginal effect of diffethnic as respondent ethnicity changes from 0 (Shi‘i) to
figure 5.14. Predicted Values of Response Variable, with Alternative Religiosity Measure
182
-1-.
50
.51
Mar
gina
l Effe
ct o
f an
Inte
r-et
hnic
Inte
rvie
w
Shi'i Sunni
Ethnicity
Marginal Effect of an Inter-ethnic Interview95% Confidence Interval
Dependent Variable: Overall Political Situation
Marginal Effect of X on Y As Z Changes
model
1 (Sunni). Once again, this change is not meant literally, and clearly the values between 0 and 1
have no substantive interpretation. The idea is simply that, as with ethnicity, there is no
single or unconditional “effect” of inter-ethnic interviews; rather, our theory tells us that the
effect on diffethnic will depend on the ethnicity of a respondent, with Shi‘is tending to give
more Sunni-like responses and Sunnis tending either to give more Shi‘i-like or more pro-state
responses.28 The marginal effect of an inter-ethnic interview on a respondent’s evaluation of
Bahrain’s political situation is given above in figure 5.15, which reveals that diffethnic is a
statistically-significant predictor of political situation only among Shi‘i respondents. We can
determine this, as before, by looking at the two bands of the 95% confidence interval: wherever
these are both below or above 0, diffethnic is significant at the 0.05 p-level. Thus we easily
see that for the entire right-hand side of the ethnicity continuum the upper band has crossed
into positive values, and that, accordingly, inter-ethnic interviewing seems here to have no
effect on Sunni respondents. Among Shi‘is, however, its impact is quite significant, being
associated with a change of about one-half of a response category in the direction of more
28
Recall that we have two plausible theories with competing expectations regarding the response of Bahraini Sunnis to inter-ethnic interviewing. See supra, 172-173.
figure 5.15. Marginal Effect of an Inter-Ethnic Interview, by Respondent Ethnicity
183
1:
Ver
y G
ood
2: G
ood
3: B
ad4:
Ver
y B
ad
Opi
nion
of O
vera
ll P
oliti
cal S
ituat
ion
Same-ethnic Inter-ethnic
Type of Interview
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
favorable opinion regarding Bahrain’s political situation. This influence is illustrated above
in Figure 5.16, which once again shows predicted values of Sunni and Shi‘i respondents as
the type of interview changes (theoretically) from a same-ethnic interview to a inter-ethnic
interview. We see that whereas the predicted responses among same-ethnic interviewees is
similar to those of Figures 5.13 and 5.14 above, for Shi‘i respondents interviewed by Sunni
fieldworkers the predicted value of political situation is, as per the marginal effect charted
in Figure 5.15, decreased by approximately one-half of a response category. This influence is
so significant indeed that the predicted responses of inter-ethnic Sunni and Shi‘i interviewees
nearly converge around Y = 2.5. Finally, notice also that the slope (marginal effect) of the line
corresponding to Sunni respondents, while not statistically significant, is slightly positive, a fact
we should bear in mind as we proceed in our analysis of other political opinions.
We move next to another independent variable of considerable theoretical interest,
religiosity. According to Hypotheses 2.2A and 2.2
B, the religiousness of a respondent, taken to
figure 5.16. Predicted Values of Response Variable, by diffethnic
184
-.5
-.25
0.2
5.5
Mar
gina
l Effe
ct o
f Res
pond
ent R
elig
iosi
ty
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: Overall Political Situation
Marginal Effect of X on Y As Z Changes
model
be a proxy for the strength of his ethnic identity, should be a significant predictor of political
opinion among both Bahraini Sunna and Shi‘a. Yet for members of the former group we
expect higher levels of religiosity to be associated with more favorable opinions of political
situation whereas among Shi‘a we expect more religiosity to be associated with less favorable
opinions. Our theory is, again, that a stronger ethnic identity should cause Sunnis to rally in
support of the Bahraini government and the larger political status quo it represents, while a
stronger Shi‘i ethnic identity should cause individuals to be more critical and less accepting
of the country’s current political order. From Figure 5.17 above we have some evidence that
this is in fact the case: we see that the marginal effect of respondent religiosity is positive
among Shi‘is and negative among Sunnis. The statistical confidence of the latter effect gives
us pause, yet the overall picture in any case seems clear: Shi‘i respondents tend to report more
critical evaluations of the political situation in Bahrain as they increase in religiosity; this
influence among Sunnis is precisely the opposite. In this, then, we have confirmation of the
pattern witnessed already in Figure 5.13, which showed that the predicted responses of Shi‘is
and Sunnis diverged as religiosity increased from 0 to 1. There we were concerned with the
figure 5.17. Marginal Effect of Religiosity, by Respondent Ethnicity
185
-.25
0.2
5
Mar
gina
l Effe
ct o
f Res
pond
ent R
elig
iosi
ty
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: Overall Political Situation
model
effect of ethnic group membership as religiosity changes; here we are interested in the utterly
opposite influence of religiosity depending on the ethnicity of a respondent. Figure 5.17 serves
to confirm statistically, that is, what Figure 5.13 suggests visually. All this may be repeated,
finally, in the case of our alternative religiosity measure, whose marginal effect is plotted in
Figure 5.18. Once again we see that religiosity is an anti-government influence among Shi‘is
and a pro-government influence among Sunnis, and that both of these effects are statistically
significant at the 0.100 p-level. An analogue to the predicted values graph of Figure 5.14, the
picture of Figure 5.18 supplies even more evidence in support of our Hypotheses 2.2A and 2.2
B.
One final element of our theoretical predictions remains, however, to be tested: the
claim of Hypothesis 2.1 that the effect of ethnicity on political opinion is equal to or greater
than that of economic satisfaction. We thus turn lastly to our analysis of this independent
variable. As with the others, we must evaluate the marginal effect of economy conditional
on respondent ethnicity, yet recall that we were uncertain theoretically whether to expect
this influence of economic well-being to differ for Sunnis and Shi‘is. We hypothesized that
its effect may in fact be stronger among Sunnis, who as a group were shown in the first section
figure 5.18. Marginal Effect of Religiosity, with Alternative Religiosity Measure
186
-.5
-.25
0.2
5.5
M
argi
nal E
ffect
of R
espo
nden
t Eco
nom
y
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Economy95% Confidence Interval
Dependent Variable: Influence of Government Policy
model
of this chapter to have a more reasonable expectation of economic benefits, of employment
benefits, in exchange for political quiescence. As we can perceive from Figure 5.19, though,
this is clearly not the case. Instead, we see that the marginal effect of economic satisfaction
on a respondent’s evaluation of the political situation in Bahrain is virtually constant across
Sunnis and Shi‘is in addition to being uniformly statistically-significant at the 0.05 p-level. In
terms of magnitude, the impact of economy is substantial, with a one-unit increase in the
indicator—say, from “good” (2) to “bad” (3) or from “bad” to “very bad” (4)—producing a
0.24-increase in political situation. The worse one’s personal economy, that is, the worse
one’s evaluation of Bahrain’s overall political situation.
Compared to that of religiosity, therefore, we must say that the impact of economic
satisfaction is greater in magnitude: Figure 5.17 shows that the marginal effect of the former is
about 0.20 for Sunnis and –0.13 for Shi‘is. But this is not the whole story. Since religiosity
is coded dichotomously, these are the marginal effects of a change in religiosity across its
entire range, i.e. from its minimum of 0 (“non-religious”) to its maximum of 1 (“religious”).
By contrast, a one-unit increase in economy corresponds to a mere 33% change over its entire
range, which goes from 1 (“very good”) to 4 (“very bad”). So for a more fair comparison of
figure 5.19. Marginal Effect of Economic Satisfaction, by Respondent Ethnicity
187
1:
Ver
y G
ood
2: G
ood
3: B
ad4:
Ver
y B
ad
Opi
nion
of O
vera
ll P
oliti
cal S
ituat
ion
1: Very Good 2: Good 3: Bad 4: Very Bad
Household Economy of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
the relative effects on political situation of economy and religiosity, we must compare
their marginal effects over an equivalent range. And since we cannot meaningfully divide the
range of the religiosity variable into smaller segments, we choose to compute the marginal
effect of economy for a transformation of ±1 standard deviation away from its observed mean
of 2.08, which is 2 × 0.606 = 1.212. So we may say that the marginal effect of economy in this
case is 0.24 × 1.212 = 0.29, or a little less than 1½-times that of religiosity among Shi‘is and
2¼-times that of religiosity among Sunnis. Interestingly, then, while the marginal effect of
economic satisfaction is the same for Sunnis and Shi‘is, its impact relative to that of individual
religiosity is indeed greater among Sunnis. This result we shall keep in mind as we proceed.
Of course, the specific prediction of Hypothesis 2.1 regards the relative influences of
economy and ethnicity rather than those of economy and religiosity. Figure 5.20, which
depicts the effects of ethnicity and economy simultaneously, shows that the two comparisons
are very different. Here, the gap between the red and black lines represents the effect of
figure 5.20. Predicted Values of Response Variable, by Economic Satisfaction
188
ethnicity as values of economy increase, while the slopes of the lines represent the effect of
economy on political situation among Sunnis and Shi‘is. Looking at the far left of the
diagram where economy = 1, we see that the red line crosses exactly at Y = 2 whereas the
black line crosses around three-quarters of the distance between Y = 2 and Y = 3. And since
these lines are parallel across the entire range of economy, we know that the marginal effect of
ethnicity on political situation is steady across all values of economy at about 0.75, or nearly
three times greater in magnitude than the marginal effect of economy as per our computation
above. As for our Hypothesis 2.1 prediction, therefore, the evidence here is in its favor: the
impact of ethnic group membership on Bahrainis’ political opinions, on their evaluations of
the nation’s overall political situation, is much greater than that of their economic satisfaction,
to say nothing of the additional, augmenting effect of individual religiosity. Accordingly, we
may conclude in this case that the individual-level causal story underlying the rentier state
framework of Arab Gulf politics, if not wrong per se, is quite incomplete.
Now that we are familiar with the method of analysis to be employed in evaluating
Hypotheses 2.1 and 2.2, for the sake of space we may move somewhat more quickly through
the eight political opinions remaining for consideration, seeing that the only new element in
each case will be the specific survey question (that is, opinion) serving as dependent variable.
Next on this list, then, is respondents’ opinion of the overall influence of government policy
on their daily lives. Bahrainis were asked to answer the question: “In general, do you feel that
government policies have an influence on your daily life?”29 with one of these five responses:
(1) “they have a very positive influence”; (2) “they have a positive influence”; (3) “they have
neither a positive nor negative influence”; (4) “they have a negative influence”; or (5) “they
have a very negative influence.” Below in Figure 5.21 we see the proportions of Sunnis and
Shi‘is that answered in each of the five categories. Overall, the picture looks little different
from that witnessed above in Figure 5.10 in the case of political situation: Bahrain’s Sunni
respondents disproportionately offer a positive or neutral evaluation of government policy,
while the tendency among Shi‘a is exactly the opposite. Indeed, a combined 81.7% of Sunni
responses fall in categories 1 through 3, whereas 80.8% of Shi‘i responses fall in categories 3
through 5. And were we to exclude the “Not Sure” and “Refuse” responses, these percentages
would be nearly 10 points higher.
29
The corresponding Arabic is: “ ؟ حياتك اليوميةیبشكل عام، هل تشعر بأن سياسات احلكومة هلا تأثري عل ”
189
VeryPositive Positive Neither Negative
VeryNegative Not Sure Refuse
shi'isunni
Per
cent
age
of R
espo
nden
ts
010
2030
40
Our model estimations below in Table 5.22 appear to tell a similar story, the ethnicity
indicator seeming to be a quite large and highly significant determinant of influence along with
diffethnic and diffethnic × ethnicity. Yet to be sure we must repeat our procedure above
and examine the marginal effects of ethnicity and diffethnic across the entire range of their
respective modifying variables. What we can say from the raw model output of Table 5.22,
however, is that in contrast to our analysis of political situation, here we find that none of
the three control variables—age, gender, or education level—is a significant predictor in any
of the three estimations, a result owing perhaps to the overwhelming impact of ethnic group
membership as illustrated in Figure 5.21. A final fact worth mentioning is that our alternative
religiosity measure—“Are you religious or not?”—seems to do a better job here of predicting
influence (in Model 2) than does our standard indicator (in Model 1). To determine whether
figure 5.21. Opinion of the Influence of Government Policy on Daily Life, by Ethnicity
190
table 5.22. Bahrainis’ Opinion of the Influence of Government Policy on their Lives, estimated three ways
Variables Model 1 Model 2 Model 3
Standard Alt. Religiosity Measure OLS Regression
B sb p >|z| B s
b p >|z| B s
b p >|t|
ethnicity -2.158 0.461 0.000 -1.756 0.449 0.000 -1.784 0.358 0.000
diffethnic -0.961 0.182 0.000 -0.965 0.188 0.000 -0.762 0.137 0.000
diffeth × eth 1.303 0.322 0.000 1.314 0.311 0.000 1.037 0.253 0.000
age -0.00462 0.00455 0.310 -0.00438 0.00446 0.326 -0.00348 0.00359 0.334
female 0.0811 0.118 0.494 0.0224 0.118 0.850 0.0630 0.0935 0.501
education 0.0346 0.0472 0.463 0.0458 0.0465 0.325 0.0290 0.0373 0.437
economy 0.267 0.151 0.078 0.252 0.155 0.105 0.190 0.112 0.092
econ × eth 0.311 0.217 0.153 0.306 0.216 0.156 0.275 0.169 0.104
religiosity 0.135 0.155 0.383 0.0960 0.0869 0.269 0.103 0.122 0.402
relig × eth -0.147 0.235 0.533 -0.366 0.129 0.004 -0.110 0.188 0.559 Constant - n/a - - n/a - 3.364 0.371 0.000
N 352 344 352
Prob. > F (χ2) 0.0000 0.0000 0.0000
(Pseudo) R2 0.1573 0.1564 0.3761
Note: Robust standard errors reported for all models
this corresponds to a substantive discrepancy, though, we must wait to compare the relative
magnitude and significance of their marginal effects, in which direction we now move.
We may start by confirming the implication of Figure 5.21 that Bahrainis’ opinion of
the salutary or detrimental character of government policy is strongly influenced by an
individual’s ethnic group membership. This we conclude from Figure 5.23 below, which
shows that the marginal effect of ethnicity is substantively large and statistically significant
across the entire range of its modifying variable religiosity. We observe further that, as in
the case of Bahrainis’ opinion of their overall political situation, this “ethnic effect” seems to
increase with a respondent’s personal religiosity: among non-religious respondents Sunni
191
-1.5
-1-.
50
.5
Mar
gina
l Effe
ct o
f Eth
nici
ty
Not Religious Religious
Religiosity of Respondent
Marginal Effect of Ethnicity95% Confidence Interval
Dependent Variable: Influence of Government Policy
Marginal Effect of X on Y As Z Changes
model
Standard Religiosity Measure Alternative Religiosity Measure
1
: Ver
y2:
Pos
itive
3: N
eith
er4:
Neg
ativ
e5:
Ver
y
Opi
nion
of t
he In
fluen
ce o
f Gov
ernm
ent P
olic
y
Non-Religious Religious
Religiosity of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
1
: Ver
y2:
Pos
itive
3: N
eith
er4:
Neg
ativ
e5:
Ver
y
Non-Religious Moderate Religious
Religiosity of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
figure 5.23. Marginal Effect of Ethnicity, by Respondent Religiosity
figure 5.24. Predicted Values of Response Variable, by Religiosity
192
ethnicity is associated with an estimated 0.97-point decrease in the dependent variable
influence, whereas among religious individuals this estimated impact is –1.08, a difference
of a bit more than 11%. In other words, the Sunni-Shi‘i discrepancy in opinion regarding the
nature of government policy appears to increase, as per our hypotheses, with religiosity.
We may offer a visual representation of this modifying effect by plotting once again
the predicted values of influence for Sunnis and Shi‘is as the religiosity variable increases.
The two panels of Figure 5.24 correspond to our two alternative measures of religiosity.
On the left side we see the predicted values of influence using our standard dichotomous
measure. Corresponding to Figure 5.23, this plot reveals the extent of the between-ethnic
difference in opinion: the predicted response of a Sunni is about 2 (that government policies
have a “positive” impact) while that of a Shi‘i is about 4 (that they have a “negative” impact).
Yet we also see that the moderating influence of religiosity, if still discernable among both
Sunnis and Shi‘is, in fact appears weak and seems to operate disproportionately among Shi‘i
respondents. That it does exist nonetheless, however, we find some evidence in the other half
of Figure 5.24, which repeats the procedure employing the alternative “Are you religious or
not?” measure of a respondent’s religiosity. On the right we perceive the basic pattern of the
left much more distinctly, with the predicted opinions of “religious” Sunnis and Shi‘is being
noticeably more extreme than their “non-religious” co-ethnics.30
To verify that this graphical pattern actually denotes statistical significance, we compute
the marginal effects of the two measures of religiosity below in Figure 5.25. We see that, in
line with our visual observations, whereas the standard religiosity variable does not predict
influence for respondents of either ethnicity at the 90% confidence level, our alternative, more
direct measure is a statistically significant predictor among Sunnis and nearly is such among
Shi‘is. While we might thus have liked to see both of our measures of religiosity perform
equally well, still the Bahrain mass survey data point overall to the same two conclusions: Shi‘a
tend to offer much more negative evaluations of Bahrain’s government and political status
quo than do Sunnis; and individuals of both ethnic groups tend to be only further entrenched
in their respective, conflicting opinions as the strength of their ethnic identification increases.
30
In fact, the moderating effect of religiosity among Shi‘is is relatively unaffected by the choice of religiosity measure; the real difference lies with Sunnis, for whom the straightforward “Are you religious or not?” indicator seems to have relatively greater validity than our standard measure. Looking back we notice the same to have been true in our foregoing analysis of the dependent variable political situation, suggesting that this peculiarity may be more general. To what extent this is indeed the case will become clearer as we proceed.
193
Standard Religiosity Measure Alternative Religiosity Measure
-.5
-.25
0.2
5.5
Mar
gina
l Effe
ct o
f Rel
igio
sity
on
Opi
nion
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: Influence of Government Policy
-.5
-.25
0.2
5.5
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: Influence of Government Policy
Still, as before we wonder about the substantive import of these two relationships—
the augmenting influence of religion as well as the more general “ethnic effect”—compared
to the impact of economic satisfaction. Is a respondent’s household economy once again a
significant predictor of political opinion in Bahrain? and, if so, is its effect overshadowed by
those of ethnicity and religion, or perhaps vice versa? The first question we may answer in
the affirmative on the basis of Figure 5.26 below, which shows that the marginal effect of the
economy variable is a statistically-significant determinant of influence along the entire range
of its modifying variable. That is to say, its marginal effect on one’s evaluation of government
policy is significant among both Sunni and Shi‘i respondents, at an estimated 0.46 among the
former group and 0.19 among the latter. For a ±1 standard deviation-change in the economy
variable, then, these effects become 0.56 and 0.23, respectively. By either measure, the impact
of economic satisfaction among Bahraini Sunnis is thus more than twice that among Shi‘is.
To answer our second question about the relative influence of economy we return to Figure
5.25, which tells us that the marginal effect of religiosity (using the alternative measure) is
about 0.07 among Shi‘is and –0.21 among Sunnis, or 0.14 and –0.42 for a full 0-to-2 change.
Thus we see that the substantive impact of economy is an estimated one and one-third times
that of religiosity among Sunnis and one and two-thirds that of religiosity among Shi‘is.
figure 5.25. Marginal Effect of Respondent Religiosity on influence, using Two Measures
194
-.5
-.25
0.2
5.5
M
argi
nal E
ffect
of R
espo
nden
t Eco
nom
y
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Economy95% Confidence Interval
Dependent Variable: Influence of Government Policy
1
: Ver
y2:
Pos
itive
3: N
eith
er4:
Neg
ativ
e5:
Ver
y
Opi
nion
of t
he In
fluen
ce o
f Gov
ernm
ent P
olic
y
1: Very Good 2: Good 3: Bad 4: Very Bad
Household Economy of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
figure 5.26. Marginal Effect of Economic Satisfaction, by Respondent Ethnicity
figure 5.27. Predicted Values of Response Variable, by Economic Satisfaction
195
-1.5
-1-.
50
.5
Mar
gina
l Effe
ct o
f Eth
nici
ty
1: Very Good 2: Good 3: Bad 4: Very Bad
Household Economy of Respondent
Marginal Effect of Ethnicity90% Confidence Interval
Dependent Variable: Influence of Government Policy
Marginal Effect of X on Y As Z Changes
But what of its impact compared to that of ethnic group membership? For this we
turn to Figure 5.27, a graph of the predicted values of influence among Sunnis and Shi‘is as
household economy changes. We easily perceive the grave implications of economic well-
being on a respondent’s opinion of Bahraini government policy: among Shi‘is a total change
in economy from “very good” to “very bad” is associated with an estimated change of about
one-half of a response category in the direction of more negative opinion; among Sunnis its
negative impact is considerably greater at a change of nearly 1.5 response categories. Yet at
the same time we see that even as household economy deteriorates, there remains a marked
between-ethnic difference in predicted response, with more than half of a response category
separating the predicted Sunni and Shi‘i answers. That respondents are in a similar position of
a “very bad” household economy, in other words, does not erase the observed Sunni-Shi‘i
discrepancy in political opinion. On the contrary, the case is precisely the opposite: the
predicted opinion of a Sunni of “very bad” economy is still more favorable than the predicted
view of government policy of a Shi‘i of “very good” economy. Another way we can visualize
figure 5.28. Marginal Effect of Ethnicity as Household Economy Changes
196
this same result is by plotting the marginal effect of ethnicity as economy changes,31 in order
to verify that the former remains a statistically-significant predictor of influence even when
economy = 4. This procedure gives us Figure 5.28 above, which shows that although the
marginal effect of ethnic group membership does indeed decrease in magnitude as household
economy deteriorates, it remains statistically-significant at an acceptable level even at the
most extreme values of economy—that is, even among the economically worst-off Bahrainis.
To conclude with a more exact comparison, we can say (according to Figure 5.24) that the
estimated marginal effect of ethnicity on influence ranges from 1.5 to 2 depending on a
respondent’s religiosity, while that of a change in economy even across its entire range—from
the best- to worst-off respondents—is only about 1.5 for Sunni respondents and 0.5 for Shi‘is.
We must conclude, therefore, that once more the substantive influence of ethnic group
membership on political opinion in Bahrain outstrips that of economic well-being, while the
latter’s impact is again greater than the augmenting effect of individual religiosity.
All that remains, therefore, is to consider our final independent variable of interest,
diffethnic, whose marginal effect in plotted below in Figure 5.29. Once again inter-ethnic
interviewing is shown to have a strong effect upon Shi‘i respondents, being associated with a
–0.75 change in the dependent variable influence. Bahraini Shi‘a who were interviewed by
Sunni fieldworkers, in other words, tended to offer less negative answers about the influence
of government policies on their lives to the tune of about three-quarters of one response
category. Among Shi‘a, then, the moderating influence of an inter-ethnic interview is greater
in magnitude even than that of a complete change in household economy from “very bad”
to “very good.” What is more, in contrast to our analysis of political situation above, here
we see that an inter-ethnic interview does not impact Bahraini Shi‘a only but Sunnis as well,
albeit to a lesser extent. According to Figure 5.29, Sunni respondents that were interviewed
by Shi‘a tend not to give more moderate or politically-safer answers as do Shi‘a of inter-ethnic
interviews, but they offer instead answers that are less moderate, more anti-government, more
akin to those likely to be given by their Shi‘a interviewers themselves. This observation is
made plain in Figure 5.30 below, which plots the predicted values of Sunni and Shi‘i responses
for single-ethnic and inter-ethnic interviews. To recall our earlier theoretical discussions about
31
Recall that until now (i.e., in Figures 5.12 and 5.23) we have been interested in the marginal effect of ethnicity as religiosity changes, which we have computed while holding economy and diffethnic constant at their means (cf. supra, note 25). Here we hold the (alternative) religiosity variable at its mean along with diffethnic.
197
-1-.
50
.5
Mar
gina
l Effe
ct o
f an
Inte
r-et
hnic
Inte
rvie
w
Shi'i Sunni
Ethnicity
Marginal Effect of an Inter-ethnic Interview90% Confidence Interval
Dependent Variable: Influence of Government Policy
1
: Ver
y2:
Pos
itive
3: N
eith
er4:
Neg
ativ
e5:
Ver
y
Opi
nion
of t
he In
fluen
ce o
f Gov
ernm
ent P
olic
y
Same-ethnic Inter-ethnic
Type of Interview
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
figure 5.29. Marginal Effect of an Inter-ethnic Interview, by Respondent Ethnicity
figure 5.30. Predicted Values of Response Variable, by diffethnic
198
the possible effects of inter-ethnic interviewing, wherein we concluded that it would be the
influence of diffethnic on Sunni respondents that would decide between the two alternative
theoretical explanations—to recall this discussion we now have an early indication as to our
answer: it seems from the positive effect of diffethnic among Sunnis that respondents of
inter-ethnic interviews moderate their preferred responses to conform not with the government
line but out of some sense of deference or shame vis-à-vis their other-ethnic interviewer. It
appears that respondents are not worried that their true responses may bring them political
trouble, that is, but that their true answers would offend or be otherwise unacceptable to the
person asking the questions, a telling portrait indeed of Sunni-Shi‘i relations in Bahrain.
The next political opinion we shall consider involves a subject that received no little
attention in our Chapter 3 analysis of Bahraini ethnic politics: the nation’s 2006 parliamentary
elections. One may recall well the ethnically- and religiously-charged atmosphere of the day,
including al-Wifäq’s decision to end its electoral boycott and the resultant founding of al-Haqq;
the political powder keg that was al-Bandar’s report on the government’s alleged program of
political naturalization, released but weeks before the vote; the timely intervention of al-Sïstänï
meant to impel Bahraini Shi‘a to the polls; and the scandal four years earlier regarding the
granting of citizenship and voting privileges to thousands of members of the al-Dawäsir and
other tribes around Saudi Arabia’s Eastern Province. At the same time, the more standard
issues of housing and employment remained a constant backdrop, forming in fact the bases
of most of the political societies’ campaigning, including that of al-Wifäq, barred as they
were from discussion of the more sensitive issues of al-tajnïs and Bandargate. What we expect
to find here, then, is empirical evidence of this influence of ethnicity and religion on Bahraini
perceptions of the 2006 election over against those of economics. Is it the case, that is, that
materially better-off individuals as a class of citizen (i.e., across the two ethnic groups) will
tend to exhibit more deference for an electoral process that, objectively-speaking, was and
remains closer to unfree and unfair than the reverse (and so will tend to preserve the socio-
economic and political status quo)? Or will we find this claimed rentier effect to be superseded
by the political agenda-setting and opinion-shaping influence of ethnic group membership
and identity? Which, in other words, is a better marker of an individual’s orientation toward
the farce of electoral politics in Bahrain: that one has a large house and fashionable car? or that
one’s house flies a black pennant and one’s car sports the sticker “ ” على حممد وآل حممداللهم صلّ
on the rear windshield?
199
Both Freeand Fair
MinorIssues
MajorIssues
Unfreeand Unfair Not Sure Refuse
shi'isunni
Fairness of ElectionsP
erce
ntag
e of
Res
pond
ents
010
2030
4050
Respondents were asked: “In general, how would you rate the freeness and fairness
of the most recent parliamentary elections that took place on November 25, 2006?”32 The
valid response options were: (1) “They were marked by complete freedom and fairness”; (2)
“They were free and fair, with some secondary problems (violations)”; (3) “They were free
and fair, with some substantial problems (violations)”; and (4) “They were neither free nor
fair.”33 Above in Figure 5.31 we find the evaluations of our respondents divided by ethnicity.
32
The Arabic is: “ ؟٢٠٠٦ نوفمرب ٢٥ جرت يف اليت األخرية بشكل عام، كيف تقيم حرية ونزاهة االنتخابات النيابية ” 33
The Arabic responses options are: “ .متيزت حبريتها ونزاهتها املطلقة .1”
“ .الثانوية) االختراقات(وجود بعض املشاكل مع كانت حرة ونزيهة، .2”
“ .اجلوهرية) االختراقات(وجود بعض املشاكل مع كانت حرة ونزيهة، .3”
“ . ونزيهةمل تكن حرة .4”
figure 5.31. Opinion of the Legitimacy of the 2006 Elections, by Ethnicity
200
As always, we perceive an extreme between-ethnic group divergence in response, with Shi‘is
tending to offer more negative opinions of the electoral process and Sunnis more positive.
Indeed, more than half of all Sunnis interviewed insisted that the elections were “completely
free and fair,” while only a combined 13% acknowledged anything more serious than minor
issues of freedom and transparency. On the other hand, just 16% of Shi‘is called the elections
“completely free and fair,” while a combined 50% cited “substantial problems” or considered
the process entirely “unfree and unfair.” The sizable number of “Not Sure” responses, finally,
is a combined product of the sensitivity of the question and the fact that many Bahrainis,
particularly Shi‘a, abstained from the vote altogether.34
When we estimate the determinants of Bahrainis’ views of the legitimacy of the 2006
election—let us call this new dependent variable election—we obtain the raw model output
given below in Table 5.32. On the surface, of course, there is relatively little we can learn
owing to the interactive nature of our independent variables of interest. Nonetheless we can
see, for example, that our estimates are stable across the three models with the exception of
that of the religiosity × ethnicity interactive term in Model 2, which utilizes our alternative
religiosity measure. This implies that our “Are you religious or not?” indicator for respondent
religiosity once again performs better among Sunnis (i.e., when ethnicity = 1) in predicting
our dependent variable than does our standard dichotomous measure. (Recall that this was
also found to be the case in our preceding analysis of influence.) If we find that this result
continues a lengthier discussion of its possible causes is probably warranted. For now, we
may proceed in our preliminary discussion of the estimation results by noting that, of our
three non-interactive control variables, only a respondent’s education level is found to be a
significant determinant of opinion about the 2006 election. Though consistent across the
three models, however, this effect is quite weak substantively: in the Model 3 linear regression
estimation, for instance, the marginal effect of a one-unit change in education level is only
about 0.08. Accordingly, even a total transformation in education from 1 (“illiterate”) to 7
(“master’s degree or higher”), a six-unit increase, would produce an estimated change in our
34
That said, a full 10 of 32 (31% of) “Not Sure” responses among Shi‘is come from those who indicate in an earlier question that they did in fact participate in the election. Among Sunnis this number is 5 of 19 (or 21%). More generally, only 22 (23%) of 96 Shi‘is and 19 (42%) of 45 Sunnis who report not voting in the election responded “I don’t know,” meaning that respondents did not systematically withhold their opinions about the legitimacy of the elections simply because they did not participate. Besides being interesting per se, this fact is important as it shows that our distribution of observations here is not truncated by a selection effect such that we only observe opinions about the election by those who voted. In such a case, we would have been forced to employ a selection model as we did in our earlier analyses of employment in Bahrain.
201
table 5.32. Bahrainis’ Opinion of the Legitimacy of the 2006 Parliamentary Elections, estimated three ways
Variables Model 1 Model 2 Model 3
Standard Alt. Religiosity Measure OLS Regression
B sb p >|z| B s
b p >|z| B s
b p >|t|
ethnicity -1.190 0.505 0.018 -0.941 0.495 0.057 -1.023 0.382 0.008
diffethnic -1.129 0.203 0.000 -1.172 0.215 0.000 -1.044 0.171 0.000
diffeth × eth 1.807 0.314 0.000 1.787 0.323 0.000 1.534 0.267 0.000
age -0.00456 0.00506 0.367 -0.00654 0.00515 0.204 -0.00362 0.00401 0.368
female 0.0363 0.138 0.792 -0.00888 0.142 0.950 0.0454 0.111 0.683
education 0.107 0.0533 0.045 0.110 0.0535 0.041 0.0799 0.0406 0.050
economy 0.141 0.127 0.268 0.188 0.135 0.164 0.112 0.115 0.328
econ × eth -0.120 0.217 0.581 -0.120 0.228 0.597 -0.108 0.165 0.512
religiosity 0.258 0.180 0.153 0.110 0.0962 0.253 0.230 0.164 0.160
relig × eth -0.250 0.281 0.373 -0.351 0.150 0.019 -0.249 0.219 0.257 Constant - n/a - - n/a - 2.278 0.428 0.000
N 340 329 340
Prob. > F (χ2) 0.0000 0.0000 0.0000
(Pseudo) R2 0.1532 0.1641 0.3597
Note: Robust standard errors reported for all models
dependent variable of just 0.48, or less than one-half of one response category in the direction
of a more cynical view. Although a statistically-significant predictor of election, therefore,
a Bahraini’s education level is hardly a substantively-significant predictor. In contrast to the
education control, notably, the other variable one might expect to be related to a respondent’s
opinion, his age, is not estimated to be a significant determinant of election at an acceptable
level of confidence, though the negative sign on the coefficient estimate is on par with our
theoretical expectations. Finally, as before and as predicted the female control continues to
be shown unrelated to Bahrainis’ political opinions.
202
-1.5
-1-.
50
.5
Mar
gina
l Effe
ct o
f Eth
nici
ty
Not Religious Religious
Religiosity of Respondent
Marginal Effect of Ethnicity95% Confidence Interval
Dependent Variable: Legitimacy of 2006 Elections
Marginal Effect of X on Y As Z Changes
model
Standard Religiosity Measure Alternative Religiosity Measure
1: F
ree/
Fai
r2:
Min
or Is
sues
3: M
ajor
Issu
es4:
Unf
ree/
fair
Opi
nion
of t
he L
egiti
mac
y of
the
2006
Ele
ctio
n
Not Religious Religious
Religiosity of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
1
: Fre
e/F
air
2: M
inor
Issu
es3:
Maj
or Is
sues
4: U
nfre
e/fa
ir
Non-Religious Moderate Religious
Religiosity of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
figure 5.33. Marginal Effect of Ethnicity, by Respondent Religiosity
figure 5.34. Predicted Values of Response Variable, by Religiosity
203
Together, Figures 5.33 and 5.34 tell exactly the same story about the impact of ethnic
group membership and religiosity as we just finished telling in our analysis of influence
above: a respondent’s ethnicity has a substantial impact on his opinion about the 2006 election,
and this influence tends to increase with individual religiosity, the latter working to amplify
this “ethnic effect” in the direction of more positive, pro-government opinion among Sunnis
and more negative opinion among Shi‘a. As there, we see according to Figure 5.33 that the
marginal effect of ethnicity is statistically-significant across the entire range of its modifying
variable religiosity, and that this effect is larger in magnitude among “religious” individuals
(an estimated –1.14) than among the “non-religious” (–0.89). We further observe that, also as
in the case of our analysis of influence (cf. Figures 5.23 and 5.24), and as noted already, our
standard measure of religiosity does relatively poorly among Sunnis in predicting respondent
opinion about the elections compared to the more direct “Are you religious or not?” indicator.
The two plots of Figure 5.34 attest to this fact: whereas the black line describing the predicted
response of Shi‘is as religiosity increases remains unchanged regardless of the measure used,
the red line describing that of Sunnis has a slope only slightly less than 0 in the left-hand plot
while in the right-hand plot using the alternative measure of religiosity we observe a strong,
negative relationship between Sunnis’ religiosity and their view of the 2006 election.
To speak more precisely, we may state on the basis of Figure 5.35 below that the
marginal effect on election of a respondent’s religiosity as measured by our standard variable
is an estimated 0.23 among Shi‘is and –0.02 among Sunnis. Whereas the former estimate is
statistically-significant at a tolerable level of confidence, however, the latter definitely is not,
so we must say that there is no discernable effect of religiosity on Sunni respondents when it
is measured by our standard indicator. Looking next to the right side of Figure 5.35, we see
that our alternative religiosity measure performs much better among Sunnis and about
equally as well as the standard measure among Shi‘a. Among the former, the marginal effect
of a one-unit change in religiosity is an estimated 0.09, among the latter –0.15. For a total
transformation in religiosity from its minimum of 0 to its maximum of 2, then, these effects
become 0.18 and –0.30, respectively, which compared to that of ethnicity are quite weak. At the
end of this analysis, therefore, we find that there is indeed an augmenting effect of religiosity
among our Bahraini survey respondents—one that causes Sunnis to be even more steadfast
in their positive evaluation of the 2006 parliamentary election and Shi‘a to be even more
critical of it—but one that also pales in comparison to the influence of ethnicity itself.
204
Standard Religiosity Measure Alternative Religiosity Measure
-.5
-.2
50
.25
.5
Mar
gina
l Effe
ct o
f Res
pond
ent R
elig
iosi
ty
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: Legitimacy of the 2006 Elections
-.
5-.
25
0.2
5.5
Mar
gina
l Effe
ct o
f Res
pond
ent R
elig
iosi
ty
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: Legitimacy of the 2006 Elections
-.5
-.25
0.2
5.5
Mar
gina
l Effe
ct o
f Res
pond
ent E
cono
my
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Economy90% Confidence Interval
Dependent Variable: Legitimacy of the 2006 Elections
figure 5.35. Marginal Effect of Respondent Religiosity on election, using Two Measures
figure 5.36. Marginal Effect of Economic Satisfaction, by Respondent Ethnicity
205
Our real interest, though, is not in the absolute impact of individual religiosity but in
its impact relative to that of economic satisfaction. That the former’s influence on election
amounts to just one-fifth or one-third of one response category may seem anticlimactic, but
if the effect of household economy is found to be less or non-existent then one’s impressions
change. And this is precisely the case here: as we glean from Figure 5.36 above, the estimated
marginal effect of economy on respondent opinion of the election is not statistically-significant
at an acceptable level of confidence at any point along the range of its modifying variable—that
is, among either Sunnis or Shi‘is.35 While this fact does not make the substantive influence of
religiosity more impressive per se, it does offer stronger support for our Hypothesis 2.1 that
ethnic group membership and individual religiousness will be more important determinants
of Bahrainis’ political opinions than economic well-being.
We have left to examine the influence of inter-ethnic interviewing on respondents’
opinions of the 2006 election. To this end Figure 5.37 below plots the marginal effect of an
inter-ethnic interview according to respondent ethnicity. We see that, as in the case of the
dependent variable influence just considered, while the influence of inter-ethnic interviewing
is statistically significant among respondents of both ethnicities, its impact is negative among
Shi‘is and positive among Sunnis. Shi‘is, that is, tend to give more Sunni-like answers when
interviewed by a member of the other ethnic group, and vice versa, though the magnitude
of this effect is about twice as large among Shi‘i interviewees. When we observe these two
effects in the predicted values plot of Figure 5.38, we are made to appreciate the implications
of inter-ethnic interviewing in Bahrain. The substantive impact of diffethnic is so immense,
in fact, that the predicted responses of Sunni and Shi‘i respondents are turned on their heads
entirely from the case of a single-ethnic interview: whereas in the latter an entire response
category separates Sunnis and Shi‘is, in the case of inter-ethnic interview respondents Shi‘is
are actually predicted to be more sanguine about the election than Sunnis, to the tune of more
than one-half of a response category. Incredibly, then, if one were to look only at inter-ethnic
interviewees one would have to assume it was indeed the Sunnis who tended to form the
government opposition in Bahrain and that the Shi‘a were its allies. Here we have further
evidence, then, in favor of the theory that inter-ethnic interviews affect respondents not by
35
At best, we might argue based on Figure 5.36 that there is some indication that economy may play a role among Shi‘i respondents in the manner predicted by rentier theorists, that lower levels of economic satisfaction are associated with less favorable opinions about the election. Yet even here, where we have not been beholden to the arbitrary p < 0.05 cut-off of “statistical significance,” such a claim would be on thin statistical ground.
206
-1.5
-1-.
50
.51
Mar
gina
l Effe
ct o
f an
Inte
r-et
hnic
Inte
rvie
w
Shi'i Sunni
Ethnicity
Marginal Effect of an Inter-ethnic Interview95% Confidence Interval
Dependent Variable: Legitimacy of the 2006 Elections
1: F
ree/
Fai
r2:
Min
or Is
sues
3: M
ajor
Issu
es4:
Unf
ree/
fair
Opi
nion
of t
he L
egiti
mac
y of
the
2006
Ele
ctio
n
Same-ethnic Inter-ethnic
Type of Interview
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
figure 5.37. Marginal Effect of an Inter-ethnic Interview, by Respondent Ethnicity
figure 5.38. Predicted Values of Response Variable, by diffethnic
207
making them less comfortable about participation in the survey more generally—in which
case those of both ethnic groups would be likely to present more pro-government views—
but by making them more mindful of their opinion vis-à-vis the opinion they ascribe to their
interviewer on the basis of the latter’s ethnicity. If this is true, then ethno-religious division
in Bahrain not only plays a critical role in shaping individual political opinion but serves also
to perpetuate this group-based political discord through its effect on basic social interaction
between ordinary Sunnis and Shi‘is, who instead of revealing their true political preferences
to the other side tend to misrepresent themselves as being less extreme and more conciliatory
in their ethnically-defined positions than they are in fact, giving the false impression that the
two factions are perhaps not as far distanced politically and socially as they nevertheless are.
The next political opinion we shall investigate is similar to the last in that it too
touches on a highly-contentious subject, and one that was likewise treated at some length in
Chapter 3. This is the debate surrounding “human rights” in Bahrain, more particularly the
Bahraini government’s respect therefor. We have seen already the way that groups such as
al-Haqq, the Bahrain Freedom Movement, the Bahrain Human Rights Society, and especially
the Bahrain Centre for Human Rights have utilized the well-developed international human
rights regime as a way of garnering outside attention for the otherwise-domestic issue of
Shi‘i-Sunni—or more accurately Shi‘i-state—relations in Bahrain. In this sense any mention
of “human rights” in Bahrain is understood immediately to mean something more akin to
“Shi‘a rights,” and Bahraini political activists do not shy in stretching this euphemism to
categories beyond those normally associated with the term. At the same time, however, the
degree of international notice directed in this way to the political situation in Bahrain has not
been lost on the government, which as discussed previously has consequently banned the
Bahrain Centre for Human Rights as of 2004 (the Bahrain Freedom Movement has operated
in exile from London since the 1990s); arrested and later pardoned its founder ‘Abd al-Hädï
al-Khawäjah on several occasions; and has established parallel, state-sponsored “human rights”
organizations such as the Bahrain Human Rights Watch Society (cf. Ch. 3, 64) that work to
counter the successful propaganda of the BCHR and other groups. Even more recently, the
government moved in September 2010 to dissolve the governing board of the oldest and
only remaining legally-operating rights organization in Bahrain, the Bahraini Human Rights
Society, appointing a new, government-backed director. The steps were taken, according to
the Ministry of Social Development, because of complaints received from Bahraini journalists
208
who “were insulted by members of the society for merely requesting the society to extend
its activities to all sections of the Bahraini community instead of limiting them to one
section”—that is, of course, the Shi‘a section. The group was accused further of “coordinating
with a number of illegal entities,” which refers again to Shi‘a opposition groups or perhaps
to the BCHR itself.36 In sum, then, the debate over “human rights” in Bahrain is an illustrative
microcosm of the larger Shi‘a-government conflict, wherein the former seeks to dictate the
terms of debate and thereby to elicit the sympathy of foreign governments, to transform the
issue from a domestic political problem to one of more universal interest and import; while
the latter hopes to silence these critics or at least to drown out their message with competing
voices, voices that tend to rebrand these “human rights campaigners” as political malcontents
—indeed “terrorists”—in sheep’s clothing. For this reason one of the most anticipated days
of the Bahraini political calendar is the release of the U.S. State Department’s annual Human
Rights Country Report for Bahrain, the relative reproof or encouragement contained in which
is taken as a tangible end-of-season scorecard for this perennial contest.
The question for our purposes thus turns around the relative influences of ethnicity,
religion, and economics in determining Bahrainis’ orientations toward this controversy. In
particular, where do individuals fall on the continuum of protecting individual rights on the
one hand and maintaining a stable society on the other? Are materially better-off Shi‘is more
willing to overlook the government record on human rights in preservation of the status quo?
Are Sunnis necessarily more supportive of the state’s conduct simply by virtue of their ethnic
rivalry with Shi‘is? Should one exist, is the intensity of this “ethnic effect” increased by one’s
individual religiosity as we have seen before? Do Sunnis and Shi‘is misrepresent their true
opinions about respecting human rights when asked by a member of the other ethnic group?
This and more we shall learn by our analysis of the following survey question: “To what
extent do you think that the lack of respect for human rights in Bahrain is justifiable in order
to maintain security?”37 Though it may seem tailor-made for the Bahraini context, this question
in fact is a standard instrument of the Arab Barometer questionnaire. But whereas it was
intended originally to measure a respondent’s relative preference for stability over freedom
36
In Amänï al-Musqatï, 2010, “ جتمد إدارة اجلمعية البحرينية حلقوق اإلنسان» التنمية «:جملس اإلدارة يقرر اللجوء إىل القضاء للطعن يف القرار ” [“The Governing Council Decides to Go to Court to Challenge the Decision: ‘[The Ministry of] Development’ Freezes the Administration of the Bahraini Society for Human Rights”], Al-Wasat, September 9. Available (in Arabic) at: <http://www.alwasatnews.com/2925/news/read/473750/1.html>. 37
The Arabic is: “ ؟ر األمن مربی علظترام حقوق اإلنسان يف البحرين للحفا أي درجة تعتقد أن عدم احیإل ”
209
High Moderate Small Not at all Not Sure Refuse
shi'isunni
Respect for Human Rights
Per
cent
age
of R
espo
nden
ts
010
2030
4050
6070
in the abstract, here it carries much more practical connotations, representing a referendum
on the Bahraini government’s very own line of argument, which is that security crackdowns
on political activists (or “terrorists” seeking and/or inciting others to “overthrow the regime”)
are necessary to uphold public security and order. Asked of Bahraini respondents in early
2009, then, amid an ongoing “security crackdown” that had seen the arrest and subsequent
royal pardon of some 178 (mostly Shi‘a) demonstrators and opposition leaders, this question
must be understood as inquiring about a very timely and very real societal issue, one that
several Shi‘i respondents admitted had touched their own families directly.
Above in Figure 5.39 we find a graph of respondent answers whose pattern comes as
perhaps little surprise: whereas Sunni Bahrainis tend to prioritize public security over respect
figure 5.39. Opinion of the Justifiability of Human Rights Violations in Bahrain, by Ethnicity
210
for human rights, responding that sacrificing the latter in the name of order is justified to a
“high” (25%) or “moderate” degree (39%), the vast majority (69%) of Shi‘is reject this notion
outright, deeming it “not at all” justified, compared to less than one-quarter of Sunnis. In
fact, the condemnation of human rights violations among the Shi‘a is so universal (i.e., there
is so little variation in the dependent variable among Shi‘i respondents), that it may prove
difficult to discover any additional statistical determinants of human rights among Shi‘is
beyond ethnic group membership. In any case, here we have a clear visual indication of the
extent of ethnic division in Bahrain regarding the proper handling of the country’s political
opponents and others whose “human rights” would be set aside in the execution of public
safety. Not included in Figure 5.39, finally, are the invalid answers of two additional Sunnis,
who responded respectively that “Human rights are highly respected in Bahrain” and that
“There are no human rights abuses in Bahrain!”
Moving on to our by-now familiar regression analysis procedure, we find below in
Table 5.40 the results of our three model estimations. On the whole these appear largely as
we have come to expect from our foregoing analyses: the first three regressors have estimated
effects that are large and statistically significant, as does the economy variable, and once more
our alternative religiosity indicator performs better among Sunni respondents in predicting
views toward human rights violation than does our standard dichotomous measure. We do,
however, observe something new in the estimated effects of our three control variables: not
only older individuals but also women are found to be more willing to sacrifice respect for
human rights in the name of security and stability. When we investigate this result further
by estimating separate regressions for our Sunni and Shi‘i sub-samples (not shown here), we
find that while the former effect operates among both Sunnis and Shi‘is, and indeed is nearly
twice as large in magnitude among Sunnis, the latter relationship involving gender obtains
only among Shi‘a, with the female control variable being associated with almost a one-half
category shift in the direction of more acceptance towards non-respect for human rights.38
With this one mystery is solved, as it is no wonder why when compared to Shi‘i males, who
as a group have a disproportionately large stake in the matter of the state’s respect for human
rights insofar as it is they whose rights are most likely to be violated, Shi‘i females will tend to
38
The estimated (by OLS regression) coefficient on the female variable in the Shi‘a sub-sample is –0.469 with an associated t-statistic of –3.39 and sample size of 201. In the Sunni sub-sample the coefficient estimate is –0.00628 with a p-value of 0.971 and N of 157.
211
table 5.40. Bahrainis’ Opinion of the Justifiability of Violating Human Rights, estimated three ways
Variables Model 1 Model 2 Model 3
Standard Alt. Religiosity Measure OLS Regression
B sb p >|z| B s
b p >|z| B s
b p >|t|
ethnicity -0.715 0.497 0.150 -0.892 0.513 0.082 -0.773 0.407 0.058
diffethnic -0.700 0.196 0.000 -0.679 0.201 0.001 -0.497 0.141 0.000
diffeth × eth 1.211 0.357 0.001 1.273 0.344 0.000 1.067 0.309 0.001
age -0.0205 0.00510 0.000 -0.0231 0.00514 0.000 -0.0153 0.00399 0.000
female -0.341 0.137 0.013 -0.324 0.138 0.019 -0.258 0.111 0.020
education 0.0394 0.0505 0.435 0.0380 0.0505 0.451 0.0459 0.0392 0.242
economy 0.338 0.168 0.044 0.316 0.178 0.076 0.214 0.0941 0.024
econ × eth -0.276 0.220 0.208 -0.171 0.229 0.456 -0.189 0.168 0.261
religiosity 0.228 0.198 0.250 0.144 0.104 0.164 0.114 0.134 0.396
relig × eth -0.379 0.271 0.162 -0.273 0.147 0.063 -0.265 0.226 0.241 Constant - n/a - - n/a - 3.485 0.384 0.000
N 371 358 371
Prob. > F (χ2) 0.0000 0.0000 0.0000
(Pseudo) R2 0.1838 0.1950 0.3847
Note: Robust standard errors reported for all models
report relatively more comprising views. As for the effect of age on Bahrainis’ opinions our
interpretation must be more equivocal, however, for we cannot ascribe the relatively lower
tolerance expressed by young people for human rights violations to their higher propensity
to be the target of these, since this fear cannot be said to operate strongly among Sunni youth.
Hence we must hypothesize that among Sunnis the preference among older respondents for
security over liberty results from a generational gap or changing priorities over time—in any
event, from an effect of aging itself rather than out of a sense of self-preservation. While we
cannot then wholly discount the idea that among Shi‘i respondents the effect of age owes
212
-1.5
-1-.
50
.5
Mar
gina
l Effe
ct o
f Eth
nici
ty
Not Religious Religious
Religiosity of Respondent
Marginal Effect of Ethnicity95% Confidence Interval
Dependent Variable: Justifiability of Human Rights Violation
Marginal Effect of X on Y As Z Changes
model
Standard Religiosity Measure Alternative Religiosity Measure
1
: Hig
h2:
Mod
erat
e3:
Sm
all
4: N
ot a
t all
Opi
nion
of t
he J
ustif
iabi
lity
of H
uman
Rig
hts
Vio
latio
n
Non-Religious Religious
Religiosity of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
1
: Hig
h2:
Mod
erat
e3:
Sm
all
4: N
ot a
t all
Non-Religious Moderate Religious
Religiosity of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
figure 5.41. Marginal Effect of Ethnicity, by Respondent Religiosity
figure 5.42. Predicted Values of Response Variable, by Religiosity
213
entirely to the same cause, it would seem naïve to believe that young Shi‘i males, many of
whom admitted incidentally that they had been involved in police altercations themselves,
were not swayed in their answers here by personal experience, or by the specter of it.
Returning now to consider the influence of our independent variables of interest, we
see from Figure 5.41 that things appear again as before: ethnic group membership is as always
a strong predictor of our dependent variable with a high degree of statistical confidence, being
associated with an estimated 0.92-unit decrease in human rights among irreligious individuals
and a 1.18-unit decrease among religious individuals. The between-ethnic discrepancy in
Bahrainis’ opinions of the justifiability of non-respect for human rights is thus an estimated
28% greater among those deemed “religious” by our normal measure as compared to those
coded “non-religious.” Looking at the right side of Figure 5.42, moreover, we see that this
augmenting effect of religiosity is even more pronounced if we measure it according to our
alternative “Are you religious or not?” indicator, which does a comparatively better job once
more of predicting respondents’ political opinions, especially those of Sunni respondents. The
respective marginal effect estimates for these two religiosity variables are depicted below in
Figure 5.43, which shows that only the alternative measure approaches statistical significance
at the 90% confidence level among both Sunni and Shi‘i respondents. Yet even if we would
have hoped for two more statistically- and substantively-significant estimates—the estimated
marginal effect is just 0.08 among Shi‘a and –0.13 among Sunna, by the alternative measure—
still we may be assured in the fact that in this our fourth political opinion examined the pattern
has remained always the same: by either measure, greater religiosity is associated among Shi‘is
with more negative political opinion, among Sunnis with more positive political opinion.
For our purposes, however, more important theoretically is the impact of ethnicity
and religiosity relative to that of household economy. And we see from Figure 5.44 below
that, in the present case, the latter is a significant determinant of Bahrainis’ views on human
rights violations only among Shi‘a respondents. It is significant not only statistically at the
0.05 p-level but also substantively, being associated with an estimated marginal effect of 0.21,
or 0.25 for a ±1 standard deviation-change in economy. Compared to that of religiosity
among Shi‘a, therefore, which by the alternative measure is an estimated 0.16 for a total
change from “non-religious” to “religious,” the marginal effect of economic satisfaction on
human rights is more than one and a half times as great. Some indication of this is given in
Figure 5.45, which plots how the predicted opinions of Sunnis and Shi‘is change as economic
214
Standard Religiosity Measure Alternative Religiosity Measure
-.5
-.25
0.2
5.5
Mar
gina
l Effe
ct o
f Res
pond
ent R
elig
iosi
ty
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: Justifiability of Human Rights Violation
-.
5-.
250
.25
.5
Mar
gina
l Effe
ct o
f Res
pond
ent R
elig
iosi
ty
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: Justifiability of Human Rights Violation
-.5
-.25
0.2
5.5
Mar
gina
l Effe
ct o
f Res
pond
ent E
cono
my
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Economy95% Confidence Interval
Dependent Variable: Justifiability of Human Rights Violation
figure 5.43. Marginal Effect of Respondent Religiosity on human rights, using Two Measures
figure 5.44. Marginal Effect of Economic Satisfaction, by Respondent Ethnicity
215
1
: Hig
h2:
Mod
erat
e3:
Sm
all
4: N
ot a
t all
Opi
nion
of t
he J
ustif
iabi
lity
of H
uman
Rig
hts
Vio
latio
n
Very Good Good Bad Very Bad
Household Economy of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
satisfaction decreases. We see that the predicted response of Shi‘is who report a “very good”
economic situation is around 3.25, while that of Shi‘is of “very bad” economy approaches
response category 4; the predicted response of Sunnis, we notice, does not vary substantially
with economy. At the same time that it demonstrates the substantive influence of economic
satisfaction among Shi‘a and its unimportance among Sunna, however, our predicted values
graph also serves to reiterate the still much more decisive impact of ethnic group membership
itself, which even among Sunnis and Shi‘is of “very good” household economy produces an
ethnic-based discrepancy in predicted response of an entire category, or a third of the whole
range of our dependent variable human rights.
We can say, then, that in Bahrain’s ongoing battle of nomenclature to decide “terrorists”
from “rights activists,” “security crackdowns” from “human rights violations,” citizens’
positions are determined foremost by their place along the Sunni-Shi‘i divide, a position that
is further bolstered if not intensely by the added impact of individual religiosity. On the other
hand, the potential economic benefit to society and citizens of increased stability owing to
figure 5.45. Predicted Values of Response Variable, by Economic Satisfaction
216
more lax enforcement of human rights (i.e., more vigorous prosecution of political opponents)
seems to play a limited systematic role in influencing individuals. Among Shi‘is, true, more
economically well-off individuals we found to be more accepting of the latter, yet in substantive
terms this effect is but one-half that of ethnicity and, moreover, one might imagine it results
primarily from the fact that it is the poorest Shi‘is that tend to feel the brunt of what one might
euphemistically call “non-respect for human rights” on the part of the Bahraini government.
Indeed, when one young Shi‘i villager was unable to follow this question as posed by the
interviewer (i.e., as it appears on the survey questionnaire—in fushä or literary Arabic), he
asked by way of clarification, “You mean when they go around beating and arresting people
[i.e., during street protests] and then they say it’s ‘to maintain security’ [‘للحفاظ علی األمن,’ as in
the original question]?” One may guess how this respondent answered.
One might also guess, in light of the divisiveness evidenced already, the likely impact
of inter-ethnic interviewing on Bahrainis’ responses to this question about human rights. In
a word, there is every reason to expect a repeat of our most recent finding, that when asked
by Shi‘is about the legitimacy of the 2006 parliamentary elections Sunnis tend to alter their
response to be even more extreme than the presumed, ethnically-ascribed view of their field
interviewer; and vice versa for Sunni respondents questioned by Shi‘is. This suspicion is
confirmed in Figures 5.46 and 5.47 below, which show, respectively, the marginal effect of
an inter-ethnic interview and the predicted responses of Sunni and Shi‘i participants in the
latter. We see that, exactly as in our foregoing analysis of election, the variable diffethnic
has such an effect on human rights that, among inter-ethnic interviewees, Shi‘is appear as
Sunnis and Sunnis as Shi‘is. More specifically, the marginal effect of diffethnic is an estimated
one-half category change in the direction of a more pro-government view among Shi‘is, and
somewhat more than this in the opposite direction among Sunni respondents. Once again,
then, the effect of Bahrain’s ethnic conflict is seen in the mass survey to operate on two distinct
levels: as a determinant, first, of individuals’ actual political opinions; and, second, of their
willingness to reveal them. And if the latter would seem more a problem for conductors of
survey research than for the country itself, one might consider whether this tendency for
self-misrepresentation in cases of between-ethnic interaction is likely to end here, or rather
to permeate more deeply and decisively into Bahraini society and politics.
As part of the standard Arab Barometer survey instrument respondents were asked to
rate their level of trust in five basic state institutions: political societies, the police, the lower
217
-1-.
50
.51
Mar
gina
l Effe
ct o
f an
Inte
r-et
hnic
Inte
rvie
w
Shi'i Sunni
Ethnicity
Marginal Effect of an Inter-ethnic Interview95% Confidence Interval
Dependent Variable: Justifiability of Human Rights Violation
1
: Hig
h2:
Mod
erat
e3:
Sm
all
4: N
ot a
t all
Opi
nion
of t
he J
ustif
iabi
lity
of H
uman
Rig
hts
Vio
latio
n
Same-ethnic Inter-ethnic
Type of Interview
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
figure 5.46. Marginal Effect of an Inter-ethnic Interview, by Respondent Ethnicity
figure 5.47. Predicted Values of Response Variable, by diffethnic
218
(elected) house of parliament, the courts, and the prime ministership. Now, if one wishes to
evoke the maximum terror and disgust in an arbitrary Bahraini, the surest way is to mention
the name of the late king’s brother, the uncle of the current king, and the prime minister since
1971, Khalïfa bin Salmän Äl Khalïfa. The longest-reigning unelected prime minister in the
world today, he is said to be the wealthiest (and, depending on whom one talks to, the most
powerful) member of the royal family, having to his name as previously mentioned the large
island of Jiddah39 and (by all accounts) the prominent dailies Al-Biläd and Akhbär al-Khalïj. One
memorable Shi‘i respondent called him “Mr. Guinness” for his place in the Guinness World
Records, another “Mr. 50-50” for his presumed cut of the national revenues. However the case,
more than any other state institution the prime ministership represents for many Bahrainis
the lack of fundamental political change, the persistence and immutability of the status quo,
in spite of the apparent reforms and improvements introduced over the previous 40 years.
While one might therefore also be interested in Bahrainis’ views toward the majlis al-nuwäb,
the courts, political societies, and so on, none of these evokes the same political symbolism
and, one imagines, polarization as the person of the prime minister.
This is not to say that this question was selected for extended analysis here simply
because it is the most likely to elicit a between-ethnic discrepancy in opinion. In fact, as we
gather from Figure 5.48 below, ethnic group membership is a critical factor in explaining
respondents’ views of each of the other institutions. In all cases, ethnic group support for an
institution tends to follow its relative power or advantage in that sphere. Thus we see that
Sunnis express demonstrably more trust in the courts, police, and the parliament itself, but
not in political parties (“societies”). Why? Because there they are relatively outmaneuvered
by al-Wifäq, which enjoys more widespread legitimacy among its constituents and high party
discipline among members. Its continuing success was evidenced most recently in the 2010
parliamentary elections, as previously related. Sunni parties, by comparison, though they
nominally form a majority caucus, in fact are split between Salafïs, Muslim Brothers, and a
more heterogeneous contingent of Äl Khalïfa tribal allies from among various family groups.
39
Indeed, the opulence of the prime minster’s residence at Jiddah was fully revealed only with the advent of the Google Earth software, which allowed ordinary Bahrainis to see for the first time the full scale of the Äl Khalïfa’s property holdings. The backlash was so great that the government immediately blocked usage of the program from inside Bahrain, only to relent some time later when it was clear that the damage had already been done. See, e.g., the commentary of one of Bahrain’s most popular bloggers: Mahmood al-Yousif, 2006, “Google Earth Does Bahrain,” Mahmood’s Den, June 12. Available at: <http://mahmood.tv/2006/06/12/google-earth-does-bahrain>.
219
A lot Moderate Little None
shi'isunni
Courts
Per
cent
age
of R
espo
nden
ts
010
2030
4050
60
A lot Moderate Little None
shi'isunni
Police
Per
cent
age
of R
espo
nden
ts
010
2030
4050
60
A lot Moderate Little None
shi'isunni
Parliament
Per
cent
age
of R
espo
nden
ts
010
2030
4050
60
A lot Moderate Little None
shi'isunni
Political Societies
Per
cent
age
of R
espo
nden
ts
010
2030
4050
60
figure 5.48. Degree of Trust in Basic State Institutions, by Ethnicity
220
That Sunni respondents should express more ambivalence about political societies than Shi‘a
is thus little wonder. In similar fashion, the relatively low trust reported for the police by
Sunni and Shi‘i alike is understandable when one recalls that although those who fill its ranks
tend of course to be Sunnis, their most distinguishable trait is not their religion but that they
are overwhelmingly “foreigners”—Pakistanis, Yemenis, Syrians, Iraqis, and so on—albeit
perhaps foreigners with Bahraini citizenship. In this way—and this indeed is precisely the
idea, or at least the intended idea—they identify neither with indigenous Bahraini Sunnis nor
Shi‘is, and vice versa, but are loyal only to the state itself.
Having thus explained the reasons for our choice of political opinion to examine next,
and shown that statistical expediency was not one of them, we may now proceed to analyze
Bahrainis’ trust in the institution of the prime ministership, and the determinants thereof.
The exact question as posed to respondents is the following: “I am going to name a group of
institutions, and I would like you to tell me to what extent you trust in each of these
institutions: / 1. The prime ministership (the prime [minister] and the [cabinet] ministers).”40
In Figure 5.49 below we find the familiar graph of the frequency of responses disaggregated
by ethnicity. We see that even compared to the considerable ethnic polarization witnessed
in the plots of Figure 5.48 above, here there is still less within-group variation in response:
60% of Sunni respondents (a full 67% of valid responses) report “a great degree” of trust in the
prime ministership, while 55% of Shi‘a (68% if we exclude refusals and “unsure” responses)
say that they have “none at all” (“ The discrepancy in the last two .(” اإلطالقیال أثق ا عل
proportions is a testament to the sensitivity of this question particularly among Shi‘a, several
of whom joked with more or less seriousness that we were trying to get them thrown in jail.
It is indeed no stretch to say that of the entire survey this is the most politically-sensitive
question asked, and for a time there was debate whether it could be asked at all.
Be that as it may, and despite the 18% who directly refused to answer and the other
13% who in reality refused by saying “I don’t know,” we succeeded nonetheless in recording
362 valid opinions about this most permanent of all modern Bahraini political institutions, a
number more than sufficient to carry out our usual statistical analysis of the relative influence
of economic well-being and ethno-religious identity in determining individuals’ answers. We
see in Table 5.50 below the results of our standard model estimation in addition to those of
40
The Arabic is: أي درجة تثق يف كل واحدة من هذه املؤسساتیموعة من املؤسسات، وأود أن ختربين إلسوف أقوم بتسمية جم : ” “ )الرئيس والوزراء(رئاسة الوزراء . ١
221
A lot Moderate Little None Not sure Refuse
shi'isunni
Respect for Human RightsP
erce
ntag
e of
Res
pond
ents
010
2030
4050
60
two new models: one restricted to Sunni respondents, another to Shi‘is. The purpose of these
new Models 2 and 3 is, in the first place, to offer a more traditional and probably more intuitive
look at the impact of our independent variables of interest, without the added complexity of
interaction terms and attendant modifying variables; but also, in the second place, to show
that in the end the latter do afford insights into the relationships we wish to understand that
are necessarily lost when we analyze our ethnic sub-samples independently, as is often done.
To begin, the most obvious and significant difference between the combined model
and the two restricted models is that the latter can never tell us the influence on our dependent
variable of ethnic group membership itself, since by definition it admits of no within-sample
figure 5.49. Degree of Trust in the Prime Ministership, by Ethnicity
222
table 5.50. Bahrainis’ Trust in the Institution of the Prime Ministership, estimated three ways
Variables Model 1 Model 2 Model 3
Standard (OLS) Sunnis Only Shi‘is Only
B sb p >|t| B s
b p >|t| B s
b p >|t|
ethnicity -1.558 0.363 0.000 - n/a - - n/a -
diffethnic -1.286 0.196 0.000 0.373 0.204 0.070 -1.288 0.200 0.000
diffeth × eth 1.633 0.278 0.000 - n/a - - n/a -
age -0.00936 0.00387 0.016 -0.00906 0.00482 0.062 -0.00919 0.00605 0.130
female 0.178 0.100 0.076 0.298 0.127 0.020 0.0679 0.154 0.660
education 0.0884 0.0377 0.020 0.0889 0.0432 0.041 0.0983 0.0643 0.128
economy 0.0948 0.106 0.370 -0.000 0.111 0.999 0.0985 0.107 0.359
econ × eth -0.0920 0.153 0.548 - n/a - - n/a -
religiosity 0.301 0.160 0.060 -0.383 0.150 0.012 0.297 0.163 0.071
relig × eth -0.660 0.219 0.003 - n/a - - n/a - Constant 3.028 0.383 0.000 1.418 0.356 0.000 3.003 0.534 0.000
N 334 149 185
Prob. > F 0.0000 0.0059 0.0000
R2 0.6160 0.1381 0.3358
Note: All three models report robust standard errors
variation in ethnicity. Thus we see that whereas the standalone ethnicity regressor has an
estimated coefficient of –1.558—corresponding to a marginal effect that is hugely significant
both substantively and statistically as depicted in Figure 5.51 below41—in our standard Model 1,
it is dropped from the Sunni- and Shi‘i-only models due to perfect collinearity. The difference,
then, is not merely procedural but substantive, and as such correspondents to substantively-
different research questions: we ask, “Among ordinary Bahraini citizens, what is the effect
on one’s trust in the prime ministership of being a Sunni rather than a Shi‘i?” The procedure
41
Note that all three models are estimated by OLS to allow straightforward comparison of marginal effects.
223
-3-2
-10
1
Mar
gina
l Effe
ct o
f Eth
nici
ty
Not Religious Religious
Religiosity of Respondent
Marginal Effect of Ethnicity95% Confidence Interval
Dependent Variable: Trust in the Prime Ministership
Marginal Effect of X on Y As Z Changes
model
of Models 2 and 3, by contrast, implies a rather different inquiry: “Among Bahraini Shi‘is (or
Sunnis), what are the factors that affect trust for the prime ministership?” Though neither is
more “correct” than the other, the first more accurately corresponds to our theoretical
interest, which lies in the individual-level influences, including crucially that of ethnicity, on
political opinion and action among Arab Gulf citizens in general, rather than the influences
among Sunnis on the one hand and Shi‘is on the other. Furthermore, we notice—according to
the reported R2 test statistics, which measure the proportion of variation in the dependent
variable accounted for in each model—that our fully-specified interactive model explains an
impressive 62% of the total variation in responses, whereas the Sunni- and Shi‘i-only models
can account for only around 14% and 34%, respectively.
At the same time as we gain an estimate of the impact of ethnicity through use of the
interactive model, moreover, we also arrive at the same place as the ethnically-segregated
models vis-à-vis our other independent variables of interest. If the apparent benefit of the Shi‘i-
and Sunni-only models is their ethnically-specific estimates of our independent variables, in
other words, then we are at no disadvantage in this regard in employing our interactive model,
which provides the same, if not immediately in the raw regression output. For example, the
figure 5.51. Marginal Effect of Ethnicity, by Respondent Religiosity
224
-.5
-.25
0.2
5.5
Mar
gina
l Effe
ct o
f Res
pond
ent R
elig
iosi
ty
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity95% Confidence Interval
Dependent Variable: Trust in the Prime Ministership
-.5
-.25
0.2
5.5
Mar
gina
l Effe
ct o
f Res
pond
ent E
cono
my
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Economy95% Confidence Interval
Dependent Variable: Trust in the Prime Ministership
figure 5.52. Marginal Effect of Respondent Religiosity, by Ethnicity
figure 5.53. Marginal Effect of Respondent Economy, by Ethnicity
225
-2-1
.5-1
-.5
0.5
1
Mar
gina
l Effe
ct o
f an
Inte
r-et
hnic
Inte
rvie
w
Shi'i Sunni
Ethnicity
Marginal Effect of an Inter-ethnic Interview95% Confidence Interval
Dependent Variable: Trust in the Prime Ministership
marginal effect of religiosity (using our standard measure) is an estimated –0.383 among
Sunnis according to Model 2 and 0.297 among Shi‘is according to Model 3; these estimates are
significant at the 0.012 and 0.071 p-levels, respectively. When we plot the marginal effect of
the religiosity variable as per our normal procedure, then, we find almost exactly the same:
Figure 5.52 shows that its impact among Shi‘is is an estimated 0.301, among Sunnis –0.359; and
this at similar levels of statistical confidence.42 Likewise, the Sunni- and Shi‘i-only models tell
us that economy has an estimated marginal effect of exactly 0.000 among Sunnis and 0.0985
among Shi‘is, but that neither estimate is statistically-distinguishable from sampling error.
Looking then at the picture of Figure 5.53 above, we find that the estimated marginal effect of
household economy according to our interactive model is a very comparable 0.00281 among
Sunnis and 0.0948 among Shi‘is and that, again, neither comes close to statistical significance
at the 95% confidence level. Finally, repeating this procedure for our last independent variable
of interest, diffethnic, we see according to our ethnically-segregated models that the estimated
42
The slight discrepancy compared to the Model 2 and 3 estimates is due to the influence of the four additional regressors that appear in the interactive model—i.e., ethnicity and the three interactive terms.
figure 5.54. Marginal Effect of an Inter-ethnic Interview, by Ethnicity
226
effect of an inter-ethnic interview on Shi‘i respondents is –1.288, on Sunni respondents 0.373,
and that both of these estimates are statistically-distinguishable from zero with a high degree
of confidence. Comparing these estimates to those of Figure 5.54 above, which charts the
marginal effect of diffethnic as ethnicity varies, we find that we have arrived once again at
almost exactly the same place, for the latter estimates the effect of diffethnic among Shi‘a as
–1.286, among Sunna as 0.347, and these at analogous levels of statistical confidence. In sum,
then, the interactive model offers us all the benefits of separate ethnic sub-sample estimations
while remedying their main drawback, namely the inability to measure and control for the
effect of ethnicity itself—for the Sunni-Shi‘i discrepancy in responses that has shown itself to
be so central thus far in determining Bahrainis’ political opinions, from their evaluations of
the country’s overall political situation to their trust in the office of the prime minister.
This methodological point having been made, we may remain here for a moment to
consider more thoroughly the substantive meaning of our findings regarding the determinants
of popular attitudes toward Bahrain’s prime minister. In the first place, we saw at the outset
that the effect of ethnic group membership is strong—indeed, the strongest of any witnessed
so far—Sunni ethnicity being associated, among “religious” individuals, with more trust in the
office of the prime minister to the tune of more than one and one-third response categories;
and among “non-religious” individuals to the extent of two entire response categories, as per
Figure 5.51. Simply being a Sunni rather than Shi‘i Bahraini, that is, produces an estimated
change from, for example, having “a great deal of trust” in the prime ministership to having
but “little trust,” this difference effected among two respondents who are otherwise identical
across all other independent variables included in our model. This great chasm separating
members of the two groups may be more easily appreciated in the familiar predicted values
plots of Figure 5.55 below, which tell a similar substantive story to that of the bar graph of
responses seen already in Figure 5.49. In the latter case, however, we had not yet controlled
for the many other factors thereafter found to influence responses, namely education, age,
female, religiosity, and diffethnic. Hence the depiction of Figure 5.55, representing as it
does the estimation results of our model rather than a mere visual depiction of descriptive
statistics, is at once more complete and more robust than the histogram of Figure 5.49.
The other thing we are made to see by the two plots of Figure 5.55 is the mediating
influence of individual religiosity on the effect of ethnic group membership. The marginal
effect of our standard religiosity measure we saw already in Figure 5.52, and as we observe
227
Standard Religiosity Measure Alternative Religiosity Measure
1
: A lo
t2:
Mod
erat
e3:
Litt
le4:
Non
e at
all
Tru
st in
the
Prim
e M
inis
ters
hip
Non-religious Religious
Religiosity of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
1: H
igh
2: M
oder
ate
3: S
mal
l4:
Not
at a
ll
Non-Religious Moderate Religious
Religiosity of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
1
: A lo
t2:
Mod
erat
e3:
Litt
le4:
Non
e at
all
Tru
st in
the
Prim
e M
inis
ters
hip
Same-ethnic Inter-ethnic
Type of Interview
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
figure 5.56 Predicted Values of prime minister, by diffethnic
figure 5.55. Predicted Values of prime minister, by Religiosity
228
above its substantive impact is comparable regardless of which measure we use, though once
again the alternative “Are you religious or not?” measure performs relatively better among
Sunnis than Shi‘is in predicting respondents’ opinions. The augmenting effect of religiosity
is, more specifically, an estimated 0.30 among Shi‘a and –0.36 among Sunna, or about a third
of a response category in each case. Here as always the effect of religiosity is not static but
varies by a respondent’s ethnic membership: for Shi‘is, being more religious corresponds to
less trust in the office of the prime minister; for Sunnis, religiosity serves to bolster one’s trust.
Finally, we may skip our customary comparison at this point of the relative substantive
influences of ethnicity, religiosity, and household economy, as we have discovered already
that the latter is a significant determinant of prime minister neither among Shi‘is nor Sunnis.
The last thing, then, that demands reiteration is the extraordinary impact of inter-
ethnic interviewing on Sunni and especially Shi‘i responses to this question. The between-
ethnic difference in survey response, so striking in Figure 5.55, is utterly erased in the case of
inter-ethnic interviewees, whose predicted response according to Figure 5.56 above is nearly
the same—at right around category 2: a “moderate degree of trust” in the prime minister—
irrespective of a respondent’s ethnicity. This result is driven primarily by the remarkable
moderating effect of diffethnic among Shi‘a, who report values of prime minister that are a
full one and a third category lower (more trusting) when asked by Sunni field interviewers.
Among Sunnis interviewed by Shi‘a this effect is precisely the opposite if substantially less
pronounced, producing a little more than a third of a category change in the direction of less
trust for the prime ministership. Quite notably, therefore, we have here the third consecutive
political opinion to have been analyzed for which the predicted responses of Sunnis and Shi‘is
are altogether inverted when obtained as part of an inter-ethnic interview.
The penultimate survey question we shall consider in this examination of the sources
of popular political opinion in Bahrain is also perhaps the most general and straightforward,
thus presenting more than any other a transparent window into the views of ordinary citizens
toward the Bahraini government. Bahraini respondents were asked to rate on a ten-point
scale their level of satisfaction with overall government performance “(أداء احلكومة).”43 The
43
In its entirety the question reads, “Using a 1-10 scale to measure the extent of your satisfaction with the performance of the Government of Bahrain, where 1 means that you are entirely unsatisfied with its performance and 10 means that you are very satisfied with its performance, ‘To what extent are you satisfied with the performance of the government?’” The corresponding Arabic reads:
أدائها، على اإلطالق عن راضٍ غري تعين أنك1رضائك على أداء حكومة البحرين، حبيث لقياس مدى 10-1 وجود مقياس من على فرض"" تعىن أنك راضٍ جداً على أدائها، إىل أي درجة أنت راضٍ عن أداء احلكومة؟10و
229
None 2 3 4 5 6 7 8 9 Total
shi'isunni
Overall Satisfaction with Gov't PerformanceP
erce
ntag
e of
Res
pond
ents
05
1015
2025
3035
relative distribution of these responses is given above in Figure 5.57. A telling picture indeed,
the near total ethnic polarization depicted in the latter requires little by way of explanation.
Whereas some 90% of Sunnis report being more satisfied than unsatisfied (i.e., report a score
of 5 or above), an almost equal proportion (82%) of Shi‘a express exactly the reverse opinion,
with a full 36% replying that they are “not at all satisfied” (“ Thus, at the .(” على اإلطالقراضٍغري
same time that a clear one-third of Bahraini Shi‘a assign the government the lowest possible
grade of overall satisfaction—with a few memorable respondents going even further to offer
such responses as “0,” “–1,” “حتت األرض” (“below the ground”; that is, less than 0), and, most
figure 5.57. Overall Satisfaction with Government Performance, by Ethnicity
230
table 5.58. Bahrainis’ Satisfaction with Overall Government Performance, estimated three ways
Variables Model 1 Model 2 Model 3
Standard (OLS) Sunnis Only Shi‘is Only
B sb p >|t| B s
b p >|t| B s
b p >|t|
ethnicity 2.285 0.793 0.004 - n/a - - n/a -
diffethnic 1.697 0.357 0.000 0.0280 0.500 0.955 1.693 0.364 0.000
diffeth × eth -1.654 0.609 0.007 - n/a - - n/a -
age 0.0230 0.00902 0.011 0.0128 0.0123 0.301 0.0326 0.0130 0.013
female -0.0956 0.247 0.699 -0.0604 0.355 0.865 -0.145 0.352 0.680
education -0.199 0.0818 0.016 -0.226 0.107 0.037 -0.167 0.126 0.185
economy -0.554 0.217 0.011 -0.0698 0.272 0.798 -0.558 0.219 0.012
econ × eth 0.463 0.336 0.170 - n/a - - n/a -
religiosity -0.997 0.331 0.003 0.558 0.379 0.142 -0.975 0.335 0.004
relig × eth 1.581 0.493 0.001 - n/a - - n/a - Constant 4.830 0.807 0.000 7.590 1.093 0.000 4.362 1.037 0.000
N 373 161 212
Prob. > F 0.0000 0.0556 0.0000
R2 0.5337 0.0675 0.2374
Note: All three models report robust standard errors
humorous of all, “Is there any choice lower than ‘1?’”—at the same time, less than one in ten
Sunnis supplies anything more negative than a neutral evaluation of government performance.
Whatever the additional influence of economic well-being, then, it is apparent already that it
will be difficult to match the immense impact of ethnic group membership.
To make this determination authoritatively, of course, we must carry out our usual
analysis, which we begin by repeating the estimation procedure of the previous dependent
variable. This procedure, one may recall, complements our standard interactive model with
separate Sunni- and Shi‘i-exclusive estimations. Though strictly speaking not as informative
231
as the interactive model, these two ethnically-segregated models offer the benefits of, first,
being more directly interpretable; and, second, allowing separate Sunni and Shi‘i estimates
of our education, female, and age control variables. The results of these three estimations
appear above in Table 5.58. We notice immediately that our independent variable estimates
are highly dependent upon a respondent’s ethnicity; in fact, of our six regressors in the Sunni-
and Shi‘i-only models, only the female control variable admits of the same substantive
interpretation—that it is unrelated to Bahrainis’ evaluations of government performance—
across both ethnic groups. As for the other two controls—education and age—we see that
the former is a significant, negative predictor of our dependent variable only among Sunni
respondents, the latter positively related to government performance only among Shi‘a.
That ethnicity is likewise a statistically-significant modifier with respect to our independent
variables of interest we can tell directly by the significance of the three interaction terms of
Model 1. Alternatively we may simply compare the coefficient and standard error estimates of
diffethnic, religiosity, and economy across the two ethnically-restricted Models 2 and 3,
which communicates the same result: household economy and participation in an inter-ethnic
interview are related to evaluations of government performance only among Shi‘is; among
Sunnis, the only variables one may connect with a high degree of confidence to government
performance is a respondent’s education level and to a lesser extent personal religiosity.
To assess the influence on Bahrainis’ views of government performance of ethnic
group membership itself, however, we must rely on our interactive model. Depicted below
in Figure 5.59 is the marginal effect of ethnicity as respondent religiosity increases. As we
have deduced already from the highly-skewed distribution of Figure 5.57 above, a respondent’s
opinion of the Bahraini government is influenced tremendously by ethnic group membership,
an effect that is powerfully conditioned in turn by the additional augmenting influence of
individual religiosity. While Sunni ethnicity is associated with more favorable views of
government performance among both “religious” and “non-religious” individuals, among the
former this effect corresponds to a between-group difference of nearly 4½ points on our 10-
point scale of responses, compared to a difference of 2.9 points among the latter category of
respondent. The effect of ethnicity on government performance is magnified by some
55%, then, among those Bahrainis who indicate that they are religious. And, as indicated by
the corresponding predicted values plots of Figure 5.60, we obtain very similar estimates of
this substantive influence of religiosity irrespective of the measure we employ.
232
-10
12
34
5
Mar
gina
l Effe
ct o
f Eth
nici
ty
Not Religious Religious
Religiosity of Respondent
Marginal Effect of Ethnicity95% Confidence Interval
Dependent Variable: Satisfaction with Overall Government Performance
Marginal Effect of X on Y As Z Changes
model
Standard Religiosity Measure Alternative Religiosity Measure
Non
e2
34
56
78
9T
otal
Leve
l of S
atis
fact
ion
with
Gov
ernm
ent P
erfo
rman
ce
Not Religious Religious
Religiosity of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
Non
e2
34
56
78
9T
otal
Not Religious Moderate Religious
Religiosity of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
figure 5.59. Marginal Effect of Ethnicity, by Respondent Religiosity
figure 5.60. Predicted Values of Response Variable, by Religiosity
233
Standard Religiosity Measure Alternative Religiosity Measure
-1.5
-1-.
50
.51
Mar
gina
l Effe
ct o
f Res
pond
ent R
elig
iosi
ty
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: Satisfaction with Overall Government Performance
-1
-.5
0.5
Mar
gina
l Effe
ct o
f Res
pond
ent R
elig
iosi
ty
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: Satisfaction with Overall Government Performance
Finally, to verify that religiosity’s role in magnifying the effect of ethnicity is not only
substantively-significant but statistically-significant as well, we compute the marginal effects
of our two alternative measures along with the 90% confidence interval of each. These we
find in Figure 5.61 above, which, even if we might have wished for greater statistical
confidence in the estimated effect of religiosity among Sunni ethnics, nonetheless serves to
reinforce the constant conclusion of our investigation thus far into the determinants of
political opinion among Bahrainis: as increased religiousness drives Shi‘is toward more anti-
government orientations, it marshals Sunnis further to the regime’s defense. To be sure, the
predicted level of satisfaction with government performance is more than 7 out of a possible
10 among religious Sunnis; among religious Shi‘is it barely reaches 3. Once more, then, we
say that the influence of economic satisfaction on government performance, the influence
of this supposed driving force behind individual political posture in rentier states, must be
powerful indeed if it is to eclipse the combined impact of ethnicity and religion.
And, in fact, as illustrated in Figures 5.62 and 5.63 below the effect of a respondent’s
household economy is not inconsiderable—not inconsiderable, that is, among Bahraini Shi‘a,
among whom the marginal effect of economy is an estimated –0.55, corresponding to a drop
figure 5.61. Marginal Effect of Respondent Religiosity, using Two Measures
234
-1-.
50
.5
Mar
gina
l Effe
ct o
f Res
pond
ent E
cono
my
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Economy95% Confidence Interval
Dependent Variable: Satisfaction with Overall Government Performance
Non
e2
34
56
78
9T
otal
Leve
l of S
atis
fact
ion
with
Gov
ernm
ent P
erfo
rman
ce
Very Good Good Bad Very Bad
Household Economy of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
figure 5.62. Marginal Effect of Economic Satisfaction, by Respondent Ethnicity
figure 5.63. Predicted Values of Response Variable, by Economic Satisfaction
235
of two-thirds of a response category in predicted satisfaction with government performance
for a ±1 standard deviation-change in economy. Yet among Sunnis the corresponding
marginal effect is not only negligible at an estimated –0.09 but also statistically-indistinguishable
from random chance, as is plain from the confidence interval bands on the right-hand side of
Figure 5.62. Of the six political opinions analyzed so far, then, this is the second instance in
which Shi‘a respondents were found to be responsive to our measure of economic satisfaction
while Sunnis were not, the other instance having been observed in the case of the dependent
variable human rights. It is worth noting that, on the other hand, the reverse case—
economy being a significant determinant of political views among Sunnis only—has not yet
been observed, a curious fact inasmuch as it would seem to contradict the logical theoretical
presumption that the bonds of political-economic patronage will tend to operate most
strongly between the Sunni-dominated Bahraini government and ordinary Sunni Bahrainis.
We will return to this observation at the end of the present section.
As always, we conclude by analyzing the impact of inter-ethnic interviewing on our
Sunni and Shi‘i respondents. Here its influence mirrors that of economy: substantively-robust
and statistically-significant among Shi‘is; non-existent among Sunnis. More precisely, Figure
5.64 below shows that the marginal effect of the diffethnic variable is an estimated 1.70 among
the former group and but 0.04 among the latter. Yet even without its usual countervailing
impact among Sunnis, the effect of an inter-ethnic interview among Shi‘a alone is strong
enough to bring the predicted responses of Bahraini Shi‘a who were interviewed by Sunnis
to within about two-thirds of a category of the predicted response of Sunnis. That is to say,
inter-ethnic interviewing produces such a positive change in Shi‘a respondents’ evaluations of
government performance that, even in the absence of the corresponding negative shift in
opinion that we typically observe among Sunnis interviewed by Shi‘is, the between-group
discrepancy in predicted opinion is cut from about two and one-third points among same-
ethnic interviewees to two-thirds of one point among inter-ethnic interviewees, or by nearly
72%. This sizable gap-bridging effect we perceive clearly in Figure 5.65.
Seemingly out of place among the preceding dependent variables, all of which have
sought to capture popular attitude toward the Bahraini regime itself, toward the fundamental
legitimacy and desirability of the nation’s prevailing political order, the final indicator to be
analyzed in this section measures the national pride of our Bahraini survey respondents. The
purpose is not to determine which of the two ethnic groups is the more patriotic—a question,
236
-1-.
50
.51
1.5
2
Mar
gina
l Effe
ct o
f an
Inte
r-et
hnic
Inte
rvie
w
Shi'i Sunni
Ethnicity
Marginal Effect of an Inter-ethnic Interview95% Confidence Interval
Dependent Variable: Satisfaction with Overall Government Performance
Non
e2
34
56
78
9T
otal
Leve
l of S
atis
fact
ion
with
Gov
ernm
ent P
erfo
rman
ce
Same-ethnic Inter-ethnic
Type of Interview
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
figure 5.64. Marginal Effect of an Inter-ethnic Interview, by Respondent Ethnicity
figure 5.65. Predicted Values of Response Variable, by diffethnic
237
incidentally, made the subject of no little debate—but to assess the extent to which the same
factors that help determine individual political opinions might also help explain the origin of
those opinions. That is, if our theory of how ethnicity and religion work to influence Bahraini
politics turns around a Sunni-Shi‘i conflict over the very nature of the state itself, one in which
the ancient, Shi‘a-governed land of Bahrän competes in the popular imagination as in the
political arena for supremacy and legitimacy over against the modern Kingdom of Bahrayn
won by “Al-Fätih” Ahmad bin Muhammad Äl Khalïfa—if this is indeed a valid conception of
the basis of socio-political division in today’s Bahrain, then we should expect to find that those
same factors that help determine one’s political standpoint are also related to the pride one
feels for being a Bahraini—for being, that is, a “true” Bahraini. Accordingly, we should find
that, contrary to our results heretofore, increased religiosity should effect the same change
in our dependent variable pride among both Sunnis and Shi‘is. Being more religious, in other
words, should not draw Sunnis and Shi‘is farther apart in opinion as we have witnessed until
now but push them in the same direction toward greater national pride, albeit toward two
separate prides for two competing conceptions of nationhood.
Respondents were asked, “How proud are you to be a Bahraini?”44 As we glean from
Figure 5.66 below, nearly four-fifths of Sunni respondents report the highest level of
national pride, compared to around 55% of Shi‘a. Two things may help account for this not
inconsiderable difference. The most obvious is a possible disproportionate inclination among
Sunna to feel (or at least convey) more patriotism toward the country and, by association,
more loyalty to the regime, an explanation that follows the trend witnessed thus far in which
Sunni ethnicity is associated with more pro-government attitude. The second likely cause is
more subtle and depends once again on the acute politicization of the standard demonym
“Bahraini,” the necessary use of which in the survey questionnaire almost certainly served to
alter the answers of some Shi‘a who choose to identify primarily as Bahärnah. Indeed, several
Shi‘i respondents specifically qualified their answer in this manner by saying, “I’m ‘very
proud’—not to be Bahraini, but to be Bahränï!” However the case, to interpret Figure 5.66
as a clear-cut comparison of national allegiance among Bahraini Sunnis and Shi‘is would
doubtless be an oversimplification.
44
The Arabic questionnaire reads: “ ؟إىل أي مدى تشعر باالعتزاز والفخر لكونك حبرينياً ”
238
Very Somewhat Little Not at all Not Sure Refuse
shi'isunni
NAtional PrideP
erce
ntag
e of
Res
pond
ents
010
2030
4050
6070
For our purposes, however, more interesting than group-level differences in response
are the determinants of national pride among individual Bahrainis, in particular the question
of the influence of personal religiosity. Do we find, as predicted, that more religious people
tend to report higher levels of national pride? and this alike among Sunnis and Shi‘is? As an
initial step toward providing an answer we may first divide our respondents between the
“religious” and the “non-religious” (according to our standard dichotomous measure) to see
whether the former group in fact seems to display higher levels of pride. The results of this
we find below in Figures 5.67 and 5.68, which seem indeed to confirm our expectation: among
religious Bahrainis—Sunnis as well as Shi‘is—the proportion of “very proud” individuals is
markedly higher compared to among religious respondents. In the case of Sunnis, the ratio of
figure 5.66. “How Proud Are You to Be a Bahraini?”; Responses by Ethnicity
239
Very Somewhat Little Not at all
shi'isunni
The Prime Ministership
Per
cent
age
of R
espo
nden
ts
020
4060
8010
0
Very Somewhat Little Not at all
shi'isunni
The Prime Ministership
Per
cent
age
of R
espo
nden
ts
020
4060
8010
0
“very proud” individuals rises from 70% among the non-religious to 85% among the religious,
an increase of more than 21%; among Shi‘is, this increase is even more dramatic, from 36%
of respondents to 62%, a change of some 72%.
To verify that this seeming relationship between personal religiosity and national
pride is not spurious, we perform our usual test of the statistical significance of the religiosity
variable as a predictor of pride. Table 5.69 below reports the estimation results of our
combined and ethnically-segregated models, which confirm that religiosity is a statistically-
significant determinant of Bahrainis’ reported levels of national pride and that, unlike in each
of our previous estimations, its effect operates in the same way on Sunnis and Shi‘is. Among
respondents of both groups, that is, identification as a religious individual is associated with
higher levels of pride. Conversely, neither of our other independent variables of interest is a
consistent predictor of pride across both ethnic groups: inter-ethnic interviewing is entirely
unrelated to respondents’ answers, while household economy is a significant predictor only
among Sunnis (though among Shi‘a its coefficient, if significant only at the 0.180 level, is of
the expected magnitude and direction). In similar fashion, each of our three control variables
is related to pride only among respondents of one of the two ethnic groups: higher levels of
education and female gender produce lower levels of pride among Shi‘a, and older Sunnis
tend to report feeling greater pride than younger Sunnis.
figure 5.67. Pride among the Non-Religious figure 5.68. Pride among the Religious
240
table 5.69. The Determinants of National Pride among Bahrainis, estimated three ways
Variables Model 1 Model 2 Model 3
Standard (OLS) Sunnis Only Shi‘is Only
B sb p >|t| B s
b p >|t| B s
b p >|t|
ethnicity -0.599 0.263 0.023 - n/a - - n/a -
diffethnic 0.133 0.113 0.243 0.00856 0.114 0.940 0.145 0.114 0.205
diffeth × eth -0.0859 0.170 0.613 - n/a - - n/a -
age -0.00447 0.00280 0.111 -0.00831 0.00398 0.039 -0.00174 0.00401 0.664
female 0.176 0.0767 0.022 0.0396 0.0941 0.675 0.258 0.116 0.027
education 0.0823 0.0323 0.011 0.0101 0.0367 0.784 0.146 0.0554 0.009
economy 0.0952 0.0910 0.296 0.182 0.0923 0.050 0.121 0.0903 0.180
econ × eth 0.0946 0.130 0.466 - n/a - - n/a -
religiosity -0.324 0.126 0.010 -0.199 0.103 0.056 -0.289 0.129 0.026
relig × eth 0.139 0.159 0.384 - n/a - - n/a - Constant 1.291 0.326 0.000 1.269 0.311 0.000 0.786 0.450 0.082
N 384 163 221
Prob. > F 0.0000 0.0781 0.0002
R2 0.1522 0.1019 0.1336
Note: All three models report robust standard errors
The marginal effects of ethnicity and religiosity we plot below in Figures 5.70 and
5.72, respectively. Together, they lend further empirical support to our above theory about
the sources of national pride among Bahrainis and, in turn, about the nature of the country’s
political conflict. As indicated already by the between-group discrepancies of Figures 5.66
through 5.68, Sunni ethnicity is shown to elicit lower values (higher levels) of pride among
both religious and non-religious individuals. Yet notably, Figure 5.70 also shows that this
influence of ethnic group membership decreases in magnitude with religiosity; that is, that
the more religious a respondent is, the lesser the impact of ethnic group membership on his
241
-1-.
50
.5
Mar
gina
l Effe
ct o
f Eth
nici
ty
Not Religious Religious
Religiosity of Respondent
Marginal Effect of Ethnicity95% Confidence Interval
Dependent Variable: Level of National Pride
Marginal Effect of X on Y As Z Changes
model
Ver
yS
omew
hat
Littl
eN
ot a
t all
How
Pro
ud A
re Y
ou to
Be
Bah
rain
i?
Non-religious Religious
Religiosity of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
figure 5.70. Marginal Effect of Ethnicity, by Respondent Religiosity
figure 5.71. Predicted Levels of National Pride, by Ethnicity and Religiosity
242
-.5
-.25
0.2
5
Mar
gina
l Effe
ct o
f Res
pond
ent R
elig
iosi
ty
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: Level of National Pride
model
answer. To state the matter still differently, as religiosity increases, it begins to eclipse ethnic
group membership as the most substantively-significant determinant of pride, a result we had
not yet even come close to witnessing in our analyses of political opinion heretofore. Among
Bahraini Sunna, for instance, the marginal effect of religiosity as given by Figure 5.72 is an
estimated –0.185, whereas that of ethnicity is an estimated –0.423 among the non-religious and
just –0.284 among the religious. When we weight the latter two effects by the proportion of
religious people in the sample, we arrive at an estimate of the substantive impact of ethnicity
that is (–0.423 × 33.2%) + (–0.284 × 66.8%) = –0.330. The substantive influence of religiosity
on national pride among Sunnis, then, is a little more than one-half the magnitude of ethnic
group membership. While thus still overshadowed by the latter effect, the influence here of
religiosity relative to that of ethnicity is the strongest of any seen so far. It is the strongest,
that is, until we consider its impact on pride among Shi‘a, which according to Figure 5.72 is
an estimated –0.324, or a mere 0.006 units shy of the weighted effect of ethnicity. Among
Shi‘i respondents, therefore, ethnic membership and personal religiosity play near-equivalent
roles in determining the amount of pride one feels for being a Bahraini, an outcome that is at
figure 5.72. Marginal Effect of Religiosity, by Ethnicity
243
once unprecedented in our analysis thus far of political opinion in Bahrain and also a useful
foreshadowing of the anticipated results to come.
For we turn now in this chapter’s final section to examine our theoretical predictions
about the determinants of political action, over and above political opinion, in Bahrain, these
contained in the two hypotheses that remain to be tested: Hypothesis 2.1A and 2.2
C. The latter,
one will remember, predict respectively that Sunni ethnicity is associated with less political
action; and that Among Sunnis as well as Shi‘is, the strength of ethnic identity [i.e., religiosity] is
positively associated with political action. The first assertion is straightforward enough; but the
second speaks directly to our investigation just completed: as with national pride, which if not
a political action per se is perhaps closer to one than to an opinion, we anticipate that the effect
of individual religiosity no longer operates in conflicting directions on Sunnis and Shi‘is but
pushes members of groups forward simultaneously. We have seen how religiosity augments
anti-government opinion among Shi‘a at the same time that it buoys pro-government opinion
among Sunna. When we reached the dependent variable pride, however, its effect diverged
from the norm, reinforcing nationalist feeling among Sunni and Shi‘i alike. Why?—because
as opinion is countered with opposite opinion, as information with misinformation, so action
is not fought with inaction but with countervailing action, with reaction. To head off a Shi‘a
takeover of parliament, to “counter probable harm,” our oft-quoted al-Ma‘äwdah implored
Bahrain’s Sunnis not to boycott the election but to vote in turn for Sunni candidates. In like
fashion, the more religious a respondent—the stronger a Bahraini’s ethnic identity—the more
likely he is to engage in political action or reaction, in the first case (as among Shi‘i opponents)
vis-à-vis the state, in the second (among ordinary Sunnis) vis-à-vis the rival faction itself. Yet
since our measures of political action cannot distinguish between these two types, we find
that personal religiosity demonstrates a similar mobilizing effect among both Sunni and Shi‘i
respondents. And indeed were this not to be the case, if our religiosity variable were found
to continue to operate in the manner encountered already, encouraging political action among
Shi‘a while stifling it among Sunnis owing to the latter’s political deference, out of a desire to
remain apolitical, then our results will have contradicted our theory. In sum, we have argued
that it is not the Shi‘a only who defy rentier assumptions in Bahrain but Sunnis as well, that the
presumed bargain of wealth-for-political quietude fails not simply due to the unique agitation
of the former, but also to the reaction of the latter. It is now time to test these claims.
244
Before moving on, however, we may spend a moment to summarize the (admittedly
extensive) results of this section in order to see how far, taken together, they bear out the
relevant predictions of our Hypotheses 2.1 and 2.2. These are: that Ethnic affiliation is equally
significant as, or more significant than, economic well-being in predicting individual political opinion
… in Bahrain (Hypothesis 2.1); where Sunni ethnicity is associated with more positive political
opinion (2.1B); and further that The strength of ethnic identity [religiosity] is a significant predictor
of individual political opinion … in Bahrain (2.2); where Among Shi‘is, stronger ethnic identity is
associated with more negative political opinion (2.2A), while Among Sunnis, stronger ethnic identity
is associated with more positive political opinion (2.2B). Now, we have already outlined our
specific model predictions, including the expected sign and significance of each independent
variable, in Table 5.9 at the outset. Though useful as a prelude to estimation, however, as a
summary of conclusions this presentation is less so, not least as it includes all interaction
terms and control variables, in whose effects we have limited direct interest. Most helpful in
evaluating the performance of our several theoretical predictions, then, will be a comparison
of the substantive effects of our four independent variables of interest—ethnicity, religiosity,
economy, and diffethnic—across the seven dependent variables we examined. Doing so gives
us Table 5.73 below, a summary of the determinants of political opinion in Bahrain.
So far we have expressed the marginal effects of our variables of interest primarily in
absolute terms: as “a 1-response category change” in some dependent variable, “a 0.75-unit
decrease,” and so on. And for the basic purpose of comparing the relative effects on the same
dependent variable of, say, religiosity and economy, this method is instructive. Now that
we wish to compare these effects across differently-scaled dependent variables, however, some
with four response categories and some with ten, we require a new nomenclature. Thus the
marginal effects of our four variables of interest are summarized below in terms of the percent
change they produce in a given dependent variable. A marginal effect of –0.75 for a question
with response categories 1 through 4, for instance, corresponds to a –0.75/3 or 25% decrease
across the entire 3-category range of the dependent variable; a marginal effect of 1.5 for a
dependent variable with 10 response categories to a 1.5/9 or 16.7% increase.
This method works without alteration for the dichotomous religiosity and diffethnic
measures. In the case of ethnicity, on the other hand, because its marginal effect depends
on an individual’s religiosity it must first be weighted as above by the proportion of religious
individuals in our sample (cf. the arithmetic of supra, 242 and note d to Figure 5.73). We must
245
similarly adjust the marginal effect of our alternative “Are you religious or not?” religiosity
measure, which is trichotomous owing to a substantial number of “moderate” respondents.
In order to maintain comparability with the dichotomous measure, therefore, we double the
estimated marginal effect so that it measures the impact of a change from a “non-religious”
person to a “religious” person, rather than to a “moderate” person. Finally, we must decide
how to calculate the substantive impact of the economy variable, whose response categories
range from “very good” to “very bad.” Here we diverge from the previous procedure, for if
we took the analogous step and multiplied the estimated marginal effect by three—i.e., if we
calculated the impact of a change from “very good” to “very bad” economy—the resulting
effect would be unrealistically exaggerated, since the observed range of the economy
variable is much more limited than that of religiosity, where many are described as both
religious and non-religious. Only 7 respondents, by contrast, reported that they had a “very
bad” household economy; indeed, more than 67% of all respondents described their economy
as “good.” To offer a more realistic measure of the substantive impact of economy, then, we
multiply its marginal effect by ±1 standard deviation of its observed mean of 2.08, which
gives us 2 × 0.606 = 1.212. In Table 5.73 below, then, the percent change in the dependent
variable is calculated using the estimated marginal effect of economy multiplied by 1.212.
This introduction given, we are ready to analyze the overall substantive influences of
our independent variables of interest on political opinion among ordinary Bahraini citizens.
In the first place, we may immediately confirm Hypothesis 2.1B that Sunni ethnicity is positively
(or, equally, that Shi‘i ethnicity is negatively) associated with more pro-government political
opinion. We easily perceive from Table 5.73 the uniformly-enormous impact of ethnic group
membership on respondents’ evaluations of the Bahraini government, wherein the positive
effect of Sunni ethnicity ranges from a low (not including the unrelated dependent variable
pride) of 26.1% to a high of 60.3%. Even the minimum observed influence of ethnicity, then,
still correspondents to a change in the associated dependent variable of more than a quarter of
its entire range of possible responses. That there exists a substantial Sunni-Shi‘i discrepancy
in Bahraini political opinion we can thus state without hesitation or qualification.
That “ethnic affiliation is equally significant as, or more significant than, economic
well-being in predicting individual political opinion … in Bahrain” (Hypothesis 2.1), we can
verify with equal confidence on the basis of Table 5.73. We see that the substantive impact
of the economy variable, where it is statistically-significant at all, is far from rivaling that of
246
table 5.73. The Determinants of Political Opinion among Bahrainis, Summary of Results
Dependent Variable (Opinion) Impact of Independent Variables of Interest
Ethnicityc Religiosity
Economy Interviewer
(Shi‘i | Sunni) (Increasing) (Worsening) (Inter-ethnic)
political situation Pro Anti | Proa Anti | Anti Pro | none
(28.6%d) (7.1%| 8.2%) (9.8%| 9.6%) (17.3%)
influence Pro Antiab
| Proa Anti | Anti Pro | Anti
b
(26.1%) (3.6%| 10.6%) (5.7%| 14.1%) (19.1%| 6.9%)
elections Pro Antib| Pro
a none | none Pro | Anti
(35.2%) (7.7%| 10.1%) (34.8%| 16.3%)
human rights Pro Antiab
| Proab
Anti | none Pro | Anti (36.5%) (5.6%| 8.5%) (8.7%) (16.6%| 19.0%)
prime minister Pro Anti | Pro none | none Pro | Anti
(60.3%) (10.0%| 12.0%) (42.9%| 11.6%)
government performance Pro Anti | Pro Anti | none Pro | none
(43.5%) (11.1%| 6.5%) (7.4%) (18.9%)
pride More More | More none | Less none | none (10.9%) (10.8%| 6.2%) (7.7%)
a Based on the alternative (“Are you religious or not?”) religiosity measure
b Coefficient has the expected sign and is significant in magnitude, but its related p-value falls slightly outside the p < 0.100
threshold of significance c Recall that since our ethnicity variable is coded 1 for Sunnis, the effect of ethnic group membership is always expressed
as the effect of a respondent’s being a Sunni rather than a Shi‘i. This choice is arbitrary of course, and to express the effect in
terms of Shi‘i ethnicity one merely has to reverse the sign of the coefficient. d The exact formula used to calculate these percentages is: [((marginal effect of ethnicity when religiosity = 0) × (1 – mean
of religiosity)) + ((effect of ethnicity when religiosity = 1) × (mean of religiosity))]/(range of dependent variable).
ethnic group membership. On the contrary, its largest substantive influence, at 14.1% in the
case of Sunnis’ opinion of the overall influence of government policy on their lives, still makes
barely one-half the impact here of ethnicity. More problematic than its magnitude, however,
is its inconsistency as a predictor of political opinion whatsoever. Even with our relatively
relaxed use of 90% as well as 95% confidence intervals to gauge the statistical significance of
marginal effects, nevertheless the economy variable was seen to be a significant predictor in
only four of our six models of political opinion, and only twice among Sunni ethnics.
In fact, then, a much better match for the substantive influence of ethnic membership
on Bahraini political opinion is the additional, augmenting effect of individual religiousness,
247
described elsewhere as the strength of ethnic identity. A proxy for the latter, our measure of
religiosity is not only a more consistent predictor of Bahrainis’ opinions than economy, but
since it operates in opposite directions on Sunni and Shi‘i respondents its aggregate divisive
impact is much more considerable. In support of Hypotheses 2.2A and 2.2
B, Table 5.73 shows
that in each of our six models of political opinion the effect of increased religiosity is to
produce more anti-government opinion among Shi‘a at the same time that it educes more
pro-government opinion among Bahraini Sunna. Although the ethnic group-specific impact
of economy does surpass that of religiosity in all but one of the cases in which the former is
a significant predictor of opinion (i.e., except in the model of government performance), still
these two sets of Sunni- and Shi‘i-conditional effects are quite comparable and, in any event,
the overall substantive impact of religiosity remains greater inasmuch as it works only to
further polarize political positions already entrenched utterly along ethnic lines.
Of all the results summarized in Table 5.73, therefore, certainly the most surprising is
this relatively weak and intermittent role of economic well-being in determining Bahrainis’,
and especially Sunni Bahrainis’, opinions. To be sure, if one were forced to choose one group
for whom economic satisfaction would surely influence opinion towards the government, it
would be the latter. If any measure of a wealth-for-acquiesce bargain exists in Bahrain, that
is, certainly it must involve the group closest ethno-religiously and, as we have demonstrated
here, politically, to the country’s leaders. Yet such seems not to be the case. Only two political
opinions investigated—Bahrainis’ evaluations of the country’s overall political situation and
of the overall influence of government policy on their lives—were related to economic
satisfaction among Sunni respondents. The remainder—their evaluations of the legitimacy
of the 2006 parliamentary elections; of the degree of respect for human rights in Bahrain; of
their trust in the prime minister; and of overall government performance—were unaffected,
whereas among Shi‘a economy influenced both human rights and government performance.
What is more, we have evidence from our Sunni- and Shi‘i-exclusive results that, quite
apart from the question of the role of economy, our empirical model of political opinion in
Bahrain preformed much better more generally in our two Shi‘i subsamples than it did in the
Sunni. Looking back, we see that the respective R2 statistics for our two ethnicity-segregated
estimations of prime minister and government satisfaction differ dramatically, at 0.336 and
0.237, respectively, for the Shi‘a sample, compared to just 0.138 and 0.0675 for the Sunni. Our
two models of Shi‘a opinion, that is, accounted for around 34% and 24% of the total variation
248
in their respective survey responses, while the Sunni-only models could explain but 14% and
7%, respectively. While some of this difference may be attributable to the somewhat larger
sample sizes of our Shi‘a-exclusive models, still one must conclude that, overall, beyond the
obvious impact of ethnicity itself we have a much more difficult time explaining the political
opinions of ordinary Bahraini Sunnis than we do those of Shi‘is.
This more general result may help elucidate our more specific finding about the lack
of influence of household economy on Sunni political opinion: it might be that the ethnic
effect among Sunnis simply tends to drown out the impact of other variables; that Sunnis as
a class of citizen represent for the state a captive ethnic constituency that is less responsive to
competing factors that might otherwise shape political opinion, for example as we observe
among Shi‘is. One might then ask why Bahraini Shi‘a, among whom the effect of ethnicity
operates equally strongly, are not then similarly immune to the influence of other variables.
This question we cannot answer but with speculation. Yet we do have some window into
the first from our interviews with (now former) Sunni parliamentarians. Three in particular
offer helpful insights: al-Asälah’s ‘Ïsä Abü al-Fath and ‘Alï Ahmad, and Sämy Qambar from
al-Manbar al-Islämï. When asked about the causes of Shi‘a frustration in Bahrain, all three
retort that, in fact, it is the Sunnis who have equal or greater cause for complaint. “[The Shi‘a]
villages used to be not cared for and were very backward and under-developed,” notes ‘Alï
Ahmad. “Now a majority of the [government’s] projects—in housing, for example—they are
targeted toward the village areas.” Sämy Qambar makes the same observation, saying,
There is an area [in my district] of al-Rifä‘ known as “Lebanon” that is one of the poorest regions of Bahrain. So it not simply that the Shi‘a are poor and Sunna rich. The challenge of poverty and socio-economic inequality [in Bahrain] is not just a Shi‘a issue or an issue based on religious differences. We [al-Manbar al-Islämï] are working in the parliament to the raise the quality of life of these people as a whole—not just Sunnis or just Shi‘a.
The most emphatic response, however, came as usual from ‘Ïsä Abü al-Fath. “Look,” he began,
the Shi‘a need to understand that none of the GCC governments pay attention to their publics—it’s not just them who are ignored. Even for us Sunnis—who represents us in the government? The Shi‘a—at least they have [Muhammad ‘Alï bin Mansür] al-Sitrï, who is a special advisor to the King [for legislative affairs]; they also have several [Shi‘i] ministers. Who do the Sunnis have? We are the ones suffering—more than them.
249
This interpretation, in fact, is not uncommon among Sunni and Shi‘i alike. It follows
an oft-heard Bahraini view that positions the royal family, and by extension the government,
not as the unwavering benefactor of Sunnis but as the final arbiter—if not an entirely neutral
one—of the island’s intrinsic social dispute. The Äl Khalïfa are in this way seen as representing
a “third ethnicity,” which in fact they are insofar as they and the other tribal families alone
follow the Mälikï tradition of Islamic jurisprudence. Even Khuri, writing in 1980 of Bahrain’s
first-ever parliamentary elections nearly a decade before, noted that
Like all other ruling families in the Gulf and Arabia, the Al-Khalifa of Bahrain consider government a legitimate right that they earned historically by defending the island against external aggression—a “right” that must not be subjected to “the fluctuating, controversial moods of public opinion,” as one Al-Khalifa sheikh put it. Members of the ruling family were not permitted to run for election because they were aloof from politics, above the National Assembly and the appeal to public opinion (225).
Much more recently, Bahrain’s Justice Minister Khälid bin ‘Alï Äl Khalïfa described the role in
similar terms, telling The Economist during pre-election turmoil in 2010 that the family positions
itself as “‘a buffer zone’ between Sunni and Shia.”45
Hence, when ‘Ïsä Abü al-Fath laments that “we Sunnis” have no one in the government
(i.e., in the executive branch) to represent “us,” he is not being disingenuous but simply
excludes as a matter of course the members of Äl Khalïfa and many other political elites from
among the allied families, who, while obviously Sunni, are above all of the ruling, tribal class,
assumed to represent ordinary Sunnis no more than does Muhammad al-Sitrï. While this is
not to endorse his claim that the Sunnis “are the ones suffering,” Abü al-Fath’s argument is
informative to the extent that it provides a possible explanation for our empirical observations.
Ordinary Sunni Bahrainis, he says in essence, are poorly-repaid for the allegiance they show
the Äl Khalïfa: disproportionately supportive of the government, they lose out on the majority
of its benefaction to the very side that opposes it. This follows of course not from malice but
from the dictates of political expediency, which say that when resources are scarce, better to
spend them where they are likely to matter most. If Sunnis can be expected to remain
supportive of the status quo, whether due to natural disposition, out of ethnic affinity, or so
as not to give political ground to their rivals, why then offer them benefits that might be used
45
Quoted in “Bahrain’s pre-election jitters. Backward steps: A Gulf monarchy’s experiment in controlled democracy is looking shaky,” 2010, The Economist, October 14. Available at: <http://www.economist.com/node/17254432>.
250
to win additional friends from among today’s enemies? When one is the Democratic nominee
for president, in other words, what use spending campaign dollars in New York?
The larger question moving forward, however, is whether this relative immunity to
the effects of economic well-being among Sunnis continues into the realm of political action.
It is one thing for individuals to offer, when asked, more positive government opinion despite
being dissatisfied economically; yet it is another to abstain from taking some action that may
help alleviate this frustration out of political deference. That ethno-religious competition drives
Sunnis to tend to support the Bahraini government irrespective of economic station, in other
words, does not preclude their acting politically on the basis of economy. On the contrary,
the complaints of the three parliamentarians summarized above seem to give the opposite
indication: that ordinary Sunnis, while certainly happy to trade the leadership of al-Wifäq or
al-Haqq for that of the Äl Khalïfa, nonetheless have grievances of their own for which they are
prepared to fight politically. Sämy Qambar even describes his role and that of his party in
these terms when he says, “We are working in the parliament to the raise the quality of life of
[the] people as a whole—not just Sunnis or just Shi‘a.” In sum, despite finding that material
well-being is an inconsistent and relatively weak predictor of Sunni political opinion in Bahrain,
we should not now assume that its impact is equally marginal in the field of political action.
A final result to underscore before proceeding is that of the influence of inter-ethnic
interviewing on respondents’ answers to survey questions. On this little more need be said:
Table 5.73 aptly summarizes the decisive impact of the diffethnic variable among Shi‘is and
Sunnis, the influence of which is surpassed only by that of ethnicity itself. Among the former
group it works in each of the six models to produce more Sunni-like, more pro-government
opinion, while the latter group it makes more anti-government, more Shi‘a-like in four of the
six cases. In these four models, moreover, the cumulative effect of an inter-ethnic interview
approaches or exceeds that even of ethnic group membership, turning reality on its head to
make the Shi‘a the relative champions of the Bahraini regime, the Sunnis now its detractors.
Two lessons, then, have we learned. The first, corresponding to our theoretical argument,
is that the political conflict between Sunni and Shi‘i in Bahrain bleeds clearly into the social
realm, altering and one must say disrupting basic inter-personal relations such that members
of the two groups are generally unwilling to reveal their true political positions to one another,
each misrepresenting himself so as to appear more in step with the views he ascribes, on the
251
basis of ethnicity, to the other. Secondly, this finding in the Bahraini context would seem to
caution care elsewhere in the selection and deployment of field interviewers, especially in
locations where rival populations are clearly segregated and where an interviewer’s preferred
survey responses may be reasonably inferred from ethnic, linguistic, or other proxies. While
such concerns are not unknown to survey research, still this problem’s particular manifestation
in societies marked by Sunni-Shi‘i division should be well heeded by those conducting such
work in the Arab and Islamic world. Presently, few if any other extant mass attitude studies
administered in the Middle East and North Africa—including those of the World Values Survey,
the Arab Barometer, and Inglehart, Moaddel, and Tessler’s (2006a; 2006b; 2008a; 2008b) surveys
of Iraq—control or even discuss the relevant characteristics of field interviewers, ethnicity or
otherwise, and thus it is impossible to know whether the transformative effect we observe
here in Bahrain operates elsewhere, and with what substantive consequences.
The Bases of Political Action in Bahrain
With such an extended preface, our analysis of political action in Bahrain we may begin with
few words of introduction. Our familiar interactive model and mode of analysis remain the
same, and we add here only one new theoretical prediction: Hypothesis 2.2C, which states that
“Among Sunnis as well as Shi‘is, the strength of ethnic identity is positively associated with
political action.” If it would thus seem like much ado to embark on a whole new investigation
simply for the sake of testing one hypothesis, recall that the latter is crucial as it can confirm
that ethnic-based political mobilization in Bahrain is not a phenomenon driven wholly by the
Shi‘a but rather entails a countervailing force composed of ordinary Sunni citizens aimed at
preventing their rivals from gaining political power and influence. So should members of
both groups, and here is Hypothesis 2.2C, exhibit more political activity and interest as their
ethnic identification, their religiosity increases: for the Shi‘a, this is directed against the state;
for Sunnis, over against the Shi‘a—in order, to say it once more, “to counter probable harm.”
The Bahrain survey contains at least five questions that may be used to measure the
extent of respondents’ involvement in politics. Indirect indicators include: (1) one’s reported
interest in politics; and (2) the extent to which one follows local political news. More direct
indicators are: whether a respondent (3) has ever signed a petition or attended a political
meeting; (4) has ever taken part in a demonstration; or (5) voted in the 2006 parliamentary
election. The first, which we call interest, corresponds to the question, “In general, what is
252
Very Interested Little Not at all Not Sure Refuse
shi'isunni
Interest in PoliticsP
erce
ntag
e of
Res
pond
ents
010
2030
40
the extent of your interest in politics?”46 We recognize in Figure 5.74 the familiar pattern of
significant Sunni-Shi‘i discrepancy in response. A combined 48% of Shi‘a report being “very
interested” or “interested” in politics, compared to just 28% of Sunnis. As always, however,
we cannot be sure of this apparent relationship between ethnicity and political interest until
we estimate our multivariate model and then the marginal effect of ethnicity.
The results of this model estimation we find below in Table 5.75, where the small
coefficient and large standard error on the ethnicity variable seem to signify on the contrary
that ethnic group membership has no independent effect on respondents’ reported levels of
political interest. In fact, apart from our control variables few indicators in the standard model
46
The Arabic is: “ ؟بصفة عامة، ما مدى اهتمامك بالسياسة ”
figure 5.74. Degree of Interest in Politics, by Ethnicity
253
table 5.75. The Determinants of Individual Political Interest in Bahrain, estimated three ways
Variables Model 1 Model 2 Model 3
Standard (OLS) Sunnis Only Shi‘is Only
B sb p >|t| B s
b p >|t| B s
b p >|t|
ethnicity -0.0757 0.353 0.830 - n/a - - n/a -
diffethnic 0.0516 0.143 0.718 -0.0939 0.227 0.680 0.0359 0.145 0.804
diffeth × eth -0.137 0.264 0.605 - n/a - - n/a -
age 0.00565 0.00371 0.129 0.00661 0.00526 0.210 0.00425 0.00528 0.422
female 0.168 0.0970 0.084 0.0489 0.141 0.729 0.293 0.133 0.028
education -0.138 0.0377 0.000 -0.0855 0.0501 0.090 -0.202 0.057 0.001
economy -0.100 0.115 0.385 0.0593 0.11 0.600 -0.123 0.116 0.287
econ × eth 0.132 0.157 0.399 - n/a - - n/a -
religiosity -0.232 0.141 0.101 -0.108 0.155 0.487 -0.259 0.142 0.069
relig × eth 0.0664 0.206 0.748 - n/a - - n/a - Constant 3.421 0.382 0.000 3.026 0.475 0.000 3.810 0.498 0.000
N 389 166 223
Prob. > F 0.0000 0.1482 0.0001
R2 0.0918 0.0506 0.1014
Note: All three models report robust standard errors
seem to be related to interest. This is especially true in the case of the Sunni-only model,
where the p-value on the F test statistic of 0.1482 means that we cannot even comfortably
reject the null hypothesis that each of our coefficients is 0—that is, that all of our independent
variables are statistically unrelated to political interest. Still, we do see that education level is
a strong, positive predictor at least among Shi‘a and probably also among Sunnis, and that
female gender is negatively related to political interest among Shi‘a.
As for the effects of our independent variables of interest, however—including that
of ethnicity—we can as usual say nothing on the basis of this raw regression output. Instead,
254
-.25
0.2
5.5
Mar
gina
l Effe
ct o
f Eth
nici
ty
Not Religious Religious
Religiosity of Respondent
Marginal Effect of Ethnicity95% Confidence Interval
Dependent Variable: Level of Political Interest
Marginal Effect of X on Y As Z Changes
model
-.5
-.25
0.2
5
Mar
gina
l Effe
ct o
f Res
pond
ent R
elig
iosi
ty
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: Level of Political Interest
model
figure 5.76. Marginal Effect of Ethnicity, by Respondent Religiosity
figure 5.77. Predicted Levels of Political Interest, by Ethnicity and Religiosity
figure 5.77. Marginal Effect of Religiosity, by Ethnicity
255
Household Economy Inter-ethnic Interview
-.25
0.2
5
Mar
gina
l Effe
ct o
f Res
pond
ent E
cono
my
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Economy90% Confidence Interval
Dependent Variable: Level of Political Interest
-.
250
.25
Mar
gina
l Effe
ct o
f an
Inte
r-et
hnic
Inte
rvie
w
Shi'i Sunni
Ethnicity
Marginal Effect of an Inter-ethnic Interview90% Confidence Interval
Dependent Variable: Level of Political Interest
Figure 5.76 shows the marginal effect of Sunni ethnicity as respondent religiosity increases
from 0 to 1. We see that, in contrast to the indications of our raw regression output, ethnic
group membership is indeed a statistically-significant determinant of political interest among
Bahrainis, but, as indicated by the two bands of the 95% confidence interval, only among
religious Bahrainis. For the latter, being a Sunni rather than a Shi‘i is associated with less
interest in politics to the extent of about 0.23 units, or a change of only about 8% across the
entire 3-unit range of the dependent variable. If still of some statistical significance, therefore,
the substantive impact of ethnicity is here not nearly so decisive as in our preceding models
of political opinion. Unlike their views about politics, it seems that Bahrainis’ interest in
politics is relatively unaffected by whether one is a Sunni or a Shi‘i.
Yet more interesting for our purposes is the influence of religiosity among Shi‘is and
Sunnis, which is charted above in Figure 5.77. Here we see that the marginal effect of the
religiosity variable on interest, as on pride before, does indeed operate in the same direction
on Sunni and Shi‘i alike: being a religious person is associated with higher levels of political
interest among members of both ethnic groups. Among Shi‘a, this effect is an estimated –0.23,
or about on par with that of ethnicity itself; among Sunnis, it is a relatively weaker (both in
figure 5.78. Marginal Effects of economy and diffethnic
256
A lot Moderate Little None Not Sure Refuse
shi'isunni
The Extent of News FollowingP
erce
ntag
e of
Res
pond
ents
010
2030
40
magnitude and in robustness of the estimate) –0.17. Thus, though preliminary and perhaps
leaving something to be desired in terms of statistical confidence, this initial result does lend
early support to our Hypothesis 2.2C about the sources of political action in Bahrain, and in turn
for our claim that ordinary Shi‘is as well as Sunnis are drawn to politics on a common basis.
As for our usual inquiry about the relative influence of religiosity compared to that of
household economy, this we may compute easily—or, rather, we need not compute it at all,
for as we see from the two graphs of Figure 5.78 above, neither a respondent’s household
economy nor inter-ethnic interviewing has any impact on a respondent’s political interest,
Sunni or Shi‘i. The only factors that seem to influence individual political interest in Bahrain
are the familiar culprits: ethnic group membership and religiosity.
figure 5.79. Following of Local Political News, by Ethnicity
257
Our second measure of political interest (and indirect measure of political action) in
Bahrain is the extent to which a respondent reports following local political news.47 As evident
from Figure 5.79 above, the pattern of Sunni and Shi‘i responses seems to resemble that just
investigated: on the whole, Shi‘is report greater following of local political news, though the
between-ethnic difference is not so manifest as it was previously in the case of Bahrainis’
political opinions. A combined 50% of Sunnis say they follow political news about Bahrain
“to a great extent” (16%) or “to a moderate extent” (34%), compared to a combined 63% of
Shi‘a (at 22% and 41%, respectively). Whether this discrepancy still exists after controlling for
our several independent variables, however, remains to be seen.
In fact, when we estimate our model of news we find that the results, reported below
in Table 5.80, continue to mirror almost exactly those of our preceding analysis of political
interest. Once more the only control variables related to one’s following of local political
news are gender and education, and these only within the Shi‘a sub-sample. Also as before,
with the exception of religiosity our independent variables of interest are generally poor
predictors of news, especially among Sunnis. Indeed, as indicated by the p-value of 0.483 on
the F test statistic reported for the Sunni-only model, the cumulative effect of our predictors
of news among Sunnis is statistically-indistinguishable from 0. Our entire Sunni-only model,
in other words, is statistically-insignificant.48
When we proceed further to estimate the marginal effects of ethnicity, religiosity,
economy, and diffethnic, the similarities between our models of interest and news continue.
First, ethnic group membership is a significant determinant of news again only for respondents
who are religious, as per Figure 5.81, and its estimated effect of 0.19 among the latter group
is very similar to the 0.23 estimate obtained in our model of political interest. Secondly and
of greater substantive importance, the marginal effect of religiosity as given by Figure 5.83
operates once more in the same direction upon both Sunni and Shi‘i respondents. This effect
is illustrated most clearly in Figure 5.82, which plots the effect of religiosity on predicted
political news following among Sunnis and Shi‘is. Rather than diverge from one another as
in all our previous plots of predicted political opinions, here the Sunni and Shi‘i lines slope in
the same direction toward increased following of local political news. Though the magnitude
47
The Arabic question reads the same: “ ؟ي مدى تتابع األخبار السياسية يف البحرينأإىل ” 48
Note that this does not preclude the possibility that our independent variables of interest are statistically-significant predictors of news among Sunnis when we estimate them from the combined interactive model.
258
table 5.80. The Determinants of Local Political News Following in Bahrain, estimated three ways
Variables Model 1 Model 2 Model 3
Standard (OLS) Sunnis Only Shi‘is Only
B sb p >|t| B s
b p >|t| B s
b p >|t|
ethnicity 0.113 0.347 0.744 - n/a - - n/a -
diffethnic -0.118 0.141 0.404 -0.0721 0.212 0.735 -0.142 0.148 0.339
diffeth × eth 0.0706 0.248 0.776 - n/a - - n/a -
age -0.00295 0.00371 0.427 -0.00974 0.00526 0.066 0.00234 0.00498 0.639
female 0.0667 0.0983 0.498 -0.0817 0.148 0.581 0.203 0.130 0.120
education -0.108 0.0400 0.007 -0.0599 0.0526 0.256 -0.175 0.0597 0.004
economy -0.00637 0.0931 0.945 -0.0257 0.105 0.808 -0.0369 0.0933 0.693
econ × eth -0.0727 0.135 0.589 - n/a - - n/a -
religiosity -0.446 0.139 0.001 -0.171 0.156 0.275 -0.467 0.139 0.001
relig × eth 0.208 0.208 0.318 - n/a - - n/a - Constant 3.243 0.386 0.000 3.302 0.476 0.000 3.428 0.508 0.000
N 389 167 222
Prob. > F 0.0154 0.4830 0.0004
R2 0.0542 0.0306 0.0999
Note: All three models report robust standard errors
of this effect is estimated to be about twice as great among Shi‘is (at 0.43) as among Sunnis
(0.23), still the upshot is the same: greater personal religiosity works among members of both
ethnic groups to augment interest in politics, in line with our theoretical expectations. What is
more, for the first time in our analysis the substantive impact of religiosity outweighs that of
ethnic group membership for all respondents. When it comes to keeping up with local politics,
more important than being a Sunni or a Shi‘i is being a religious Sunni or a religious Shi‘i.
The final commonality linking our models of interest and news is the utter lack of
explanatory power exhibited by our remaining independent variables of interest, household
259
-.25
0.2
5.5
Mar
gina
l Effe
ct o
f Eth
nici
ty
Not Religious Religious
Religiosity of Respondent
Marginal Effect of Ethnicity90% Confidence Interval
Dependent Variable: Following of Local Political News
Marginal Effect of X on Y As Z Changes
model
Larg
eM
oder
ate
Littl
eN
ot a
t all
Ext
ent o
f Kee
ping
up
with
Loc
al P
oliti
cal N
ews
Non-religious Religious
Religiosity of Respondent
Predicted Y-values if Sunni
Predicted Y-values if Shi'i
figure 5.81. Marginal Effect of Ethnicity, by Respondent Religiosity
figure 5.82. Predicted Levels of News Following, by Ethnicity and Religiosity
260
-.5
-.25
0.2
5
Mar
gina
l Effe
ct o
f Res
pond
ent R
elig
iosi
ty
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: Following of Local Political News
model
Household Economy Inter-ethnic Interview
-.25
0.2
5
Mar
gina
l Effe
ct o
f Res
pond
ent E
cono
my
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Economy90% Confidence Interval
Dependent Variable: Following of Local Political News
-.2
50
.25
Mar
gina
l Effe
ct o
f Res
pond
ent E
cono
my
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Economy90% Confidence Interval
Dependent Variable: Following of Local Political News
figure 5.83. Marginal Effect of Religiosity on News Following, by Ethnicity
figure 5.84. Marginal Effects of economy and diffethnic
261
economy and inter-ethnic interviewing, whose respective marginal effects are plotted above
in Figure 5.84. The picture is self-explanatory: neither variable approaches statistical significance
among respondents of either ethnicity, and in any case their estimated effects are negligible
in substantive terms compared to the strong influences of ethnicity and especially religiosity.
Having thus seen and compared the determinants of the dependent variables news and
interest, we cannot but conclude that these two questions measure one and the same quality
in our Bahraini survey respondents, namely the extent of one’s political engagement. Yet
political engagement is still short of political action per se, and these two preliminary results,
if seemingly in support of Hypothesis 2.2C, are not entirely satisfying. So we proceed now to
assess the relative impacts of ethnicity and religiosity over against that of economic satisfaction
on Bahrainis’ political actions proper, namely on their propensity to participate in political
meetings/petitioning, demonstrations, and elections. As each of these dependent variables
is binary, our model design changes accordingly, and we employ probit regression to estimate
the effects of our independent variables on the probability that a respondent partakes in the
three political actions of interest here.
The first indicator to be considered is constructed on the basis of the survey question:
“Here we have a group of activities that citizens sometimes engage in. In the previous three
years, have you: attended a meeting to discuss some issue or to sign a petition?”49 Respondents
could answer that they participated “one time,” “more than once,” or “not at all.” However,
to allow more straightforward interpretation and to help correct for possible underreporting
due to the question’s sensitivity, the former two responses are here collapsed into a single
category, producing a dichotomous indicator measuring participation versus non-participation.
The distribution of this variable we find in Figure 5.85 below by ethnic group. Shi‘i Bahrainis,
we see, are split nearly equally between those who report having attended a meeting/signed
a petition (49%) and those who have not (50%). Sunnis, by contrast, are much more likely to
report having not participated (78%). It would seem, then, that ethnic group membership is
strongly related to this first measure of political action.
When we look at the results of our probit model estimation in Table 5.86, we find
further that several other of our independent variables, including all three control variables,
are highly significant predictors of the meeting/petition variable. In the first place, the single
49
In Arabic: ا املواطنون، خالل السنوات الثالث املاضية، هل قمت باملشاركة يفهنالك جمموعة من النشاطات اليت عادة ما يقوم : ”
“ ؟حضور لقاء أو اجتماع من أجل حبث موضوع ما أو التوقيع على عريضة .1
262
No Yes Not Sure Refuse
shi'isunni
PetitionP
erce
ntag
e of
Res
pond
ents
020
4060
80
most important determinant in substantive and statistical terms is the female control, and
this among both Sunnis and Shi‘is as we see from the disaggregated Model 2 and Model 3 results.
More precisely, the predicted probability that a respondent reports attending a meeting or
signing a petition drops from an estimated 48% among male Bahrainis to just 19% among
females, keeping all other factors constant. Though this result is perhaps unsurprising given
social realities in the Arab Gulf, still the extent of this gender effect is impressive nonetheless.
Perhaps equally important is the influence of a respondent’s age, with younger Bahrainis
much more likely to report having taken part than older individuals. The predicted
probability of participation for a 20 year-old respondent, for example, is 46%, compared to
just 28% for a 50 year-old. If we consider only male respondents, moreover, the influence of
figure 5.85. Meeting Attendance / Petition Signing, by Ethnicity
263
table 5.86. The Determinants of Meeting Attendance/Petition Signing in Bahrain, estimated three ways
Variables Model 1 Model 2 Model 3
Standard (Probit) Sunnis Only Shi‘is Only
B sb p >|z| B s
b p >|z| B s
b p >|z|
ethnicity 0.121 0.560 0.829 - n/a - - n/a -
diffethnic 0.514 0.208 0.013 -0.373 0.326 0.253 0.521 0.209 0.013
diffeth × eth -0.879 0.378 0.020 - n/a - - n/a -
age -0.0158 0.00569 0.006 -0.0194 0.00946 0.041 -0.0137 0.00720 0.057
female -0.819 0.157 0.000 -0.874 0.254 0.001 -0.797 0.201 0.000
education 0.126 0.0607 0.038 0.0884 0.0933 0.343 0.152 0.0812 0.061
economy 0.0831 0.142 0.559 -0.128 0.199 0.522 0.0922 0.143 0.519
econ × eth -0.210 0.240 0.381 - n/a - - n/a -
religiosity 0.677 0.209 0.001 0.397 0.270 0.141 0.692 0.212 0.001
relig × eth -0.265 0.323 0.412 - n/a - - n/a - Constant -0.583 0.576 0.312 -0.118 0.872 0.893 -0.818 0.707 0.247
N 388 167 221
Prob. > F 0.0000 0.0007 0.0000
Pseudo R2 0.1929 0.1341 0.1309
Note: All three models report robust standard errors
age increases further such that these proportions become 57% and 39%, respectively. Finally,
higher levels of education are associated with higher probabilities of petition signing and
meeting attendance among respondents of both ethnic groups, though this estimated effect is
only statistically-significant within the Shi‘i subsample. The predicted probability of answering
“Yes” to meeting/petition is an estimated 47% among Shi‘i males with a secondary education,
for example, compared to a somewhat higher 59% among those having attended university,
all else being equal.
264
-.25
0.2
5
Mar
gina
l Effe
ct o
f Eth
nici
ty
Not Religious Religious
Religiosity of Respondent
Marginal Effect of Ethnicity95% Confidence Interval
Dependent Variable: Meeting Attendance/Petition Signing
Marginal Effect of X on Y As Z Changes
model
To confirm the significance of ethnic group membership we estimate the marginal
effect of ethnicity on meeting/petition as religiosity increases.50 Contrary to the raw probit
output reported above, Figure 5.87 shows that ethnic group membership does indeed affect
the likelihood of Bahrainis’ participation, both among religious and non-religious individuals.
Even more notable is the impact of personal religiosity, here as strong as ever. Its respective
marginal effects among Shi‘is and Sunnis are given below in Figure 5.88: an estimated 0.23
among the former and 0.14 among the latter, both of which are statistically-significant at the
90% confidence level. To help interpret the substantive impact of these effects we chart the
change in predicted probability of meeting attendance/petition signing resulting from the
influence of religiosity.51 These predicted probabilities we find in Figures 5.89 and 5.90 below.
50
As before, I follow the suggestion of Brambor, Clark, and Golder (2006) regarding the computation and presentation of marginal effects for multiplicative interaction models. In this case I have adapted their Stata code for binary dependent variables, available at http://homepages.nyu.edu/~mrg217/interaction.html. 51
These predicted probabilities are computed using the post-regression utility mfx in Stata, which calculates the predicted probability that a dependent variable equals 1 given specific values of independent variables. The probabilities reported in Figures 5.89 and 5.90 represent all four permutations of the ethnicity and religiosity variables (i.e., Sunni religious, Sunni non-religious, Shi‘i religious, Shi‘i non-religious). All control variables are held constant at their means, while interactive terms are evaluated as appropriate (e.g., at 0 for Shi‘i ethnics).
figure 5.87. Marginal Effect of Ethnicity, by Respondent Religiosity
265
-.25
0.2
5
Mar
gina
l Effe
ct o
f Res
pond
ent R
elig
iosi
ty
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: Meeting Attendance/Petition Signing
Marginal Effect of X on Y As Z Changes
model
The first, whose calculations are based on male and female respondents, shows that while
the marginal effects of religiosity are seemingly small in absolute terms, their relative
influences on respondent behavior are profound. For Sunnis, the estimated probability that a
non-religious respondent answers “Yes” to meeting/petition is approximately 13%, compared
to nearly 23% among religious individuals, using the Model 2 results. For Shi‘is the influence
is even more pronounced: as per Model 3, among those identified as non-religious the predicted
probability of participation is just 33%, while among the religious it is 60%. Being a religious
rather than irreligious person thus increases the likelihood of meeting attendance/petition
signing among Sunnis by some 77%, among Shi‘is by an even greater 82%.
Finally, given our finding that females are much less likely to participate than males,
we may wish to compute a more real-world estimate of the substantive impact of religiosity
on meeting/petition by assessing its effect on male respondents only. This we have in Figure
5.90, which shows, as expected, increased probabilities of participation across all types of
respondent. The likelihood that a non-religious, Sunni male will have taken part in this
political action is estimated at 23%, compared to 37% among the religious. Among Shi‘i males
figure 5.88. Marginal Effect of Religiosity, by Ethnicity
266
Predicted Petition by Sect
0
10
20
30
40
50
60
70
Sunni Shi‘i
Pre
dict
ed P
roba
bilit
y o
f P
artic
ipat
ion
(%)
Not Religious Religious
Predicted Petition by Sect
0
10
20
30
40
50
60
70
Sunni Shi‘i
Pre
dict
ed P
roba
bilit
y o
f P
artic
ipat
ion
(%)
Not Religious Religious
figure 5.89. Predicted Probability of Meeting Attendance/Petition Signing
figure 5.90. Predicted Probability of Meeting Attendance/Petition Signing, Males Only
267
these estimated probabilities are 41% and 68%, respectively. The impact of being a religious
person among Sunni males is therefore an increased probability of political participation to the
tune of some 61%; among Shi‘a respondents this reaches 66%. If the substantive influence of
heightened religiosity is somewhat reduced, then, when we restrict our analysis only to male
respondents, still its effect is tremendous, particularly among Bahraini Shi‘a.
It remains only to evaluate the effect on meeting/petition of household economy,
and the latter’s substantive significance compared to that of ethnicity and religiosity witnessed
already. Yet once more this exercise proves trivial; the left panel of Figure 5.91 shows that
the variable economy has no predictive power in our model of meeting attendance/petition
signing in Bahrain. Not only is its estimated marginal effect not significant from a statistical
standpoint, but the estimate itself is minuscule for both Sunni and Shi‘i respondents, in each
case little different from 0. Thus are our theoretical predictions about the determinants of
political action among Bahraini citizens borne out here again: Shi‘i ethnicity is associated with
a higher probability of acting; being a religious Sunni or religious Shi‘i greatly augments this
likelihood, by approximately 66% and 75%, respectively; while household economy is entirely
unrelated to a respondent’s chances of participating. So too, accordingly, do we have evidence
supporting our larger argument about the bases of political mobilization in Bahrain, that
engagement in politics, if perhaps disproportionately high among Shi‘is, is not exclusive to
them qua opposition. Sunnis, incited by ethno-religious rivalry, can be political animals too.
Before moving on to our next measure of political action—participation in a march
or demonstration—we may attempt some explanation of the right-hand panel of Figure 5.91,
which plots the marginal effect of an inter-ethnic interview for Sunni and Shi‘i respondents.
We say “attempt” because its impact in this case runs counter to previous findings, in which
inter-ethnic interviewing had always the effect of making Shi‘i respondents appear more like
Sunnis (i.e., in their political opinions) and vice versa. Here, by contrast, Shi‘is who were
interviewed by Sunnis are much more likely to report having attended a political meeting or
signed a petition than those interviewed by a co-ethnic. And though we have less statistical
confidence in the estimate of the reverse effect—that is, the impact on Sunnis interviewed by
Shi‘is—there is some indication that the diffethnic variable operates here as well, making
Sunnis less likely to answer “Yes” to having participated. It thus seems that respondents—
Shi‘i respondents, at the very least—tend to exaggerate the extent of their participation or non-
participation when asked by a member of the other ethnic group. What explains this change?
268
Household Economy Inter-ethnic Interview
-.25
0.2
5
Mar
gina
l Effe
ct o
f Res
pond
ent E
con
omy
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Economy (+/- 1 STD)90% Confidence Interval
Dependent Variable: Meeting Attendance/Petition Signing
Marginal Effect of X on Y As Z Changes
-.
250
.25
Mar
gina
l Effe
ct o
f an
Inte
r-et
hnic
Inte
rvie
w
Shi'i Sunni
Ethnicity
Marginal Effect of an Inter-ethnic Interview90% Confidence Interval
Dependent Variable: Meeting Attendance/Petition Signing
Marginal Effect of X on Y As Z Changes
Our first thought may be that the sensitivity of the question, to the extent that it asks
respondents to reveal behavior that may be supposed illegal or at least undesirable, perhaps
injects a systematic bias into our responses: respondents simply may have been untruthful in
a manner that relates to interviewer ethnicity. For this explanation to make sense, of course,
we must assume, to take the first case, that a Shi‘i’s being interviewed by another Shi‘i
somehow induces more anxiety than if he were questioned by a Sunni. All the same, for the
sake of the argument let us make this assumption. We will reason, “To be asked about political
opinions and general evaluations of the government is one thing, about actual behavior that
may land one under suspicion or arrest quite another. If a Sunni interviewer makes it more
uncomfortable for a Shi‘i to reveal his true opinions because he knows well that the former
disagrees, yet to be asked about one’s political actions by a fellow Shi‘i is just plain suspicious.”
And so it is, perhaps, that we arrive at our finding.
But this interpretation has at least two flaws. First, from a theoretical standpoint it
cannot explain the reverse case, explain why Sunnis seem to underplay their participation to
Shi‘i interviewers rather than, as one would predict on the basis of this new explanation, to
Sunnis. While strictly-speaking our estimate here of the effect of inter-ethnic interviewing is
figure 5.91. Marginal Effects of economy and diffethnic
269
-.25
0.2
5
Mar
gina
l Effe
ct o
f an
Inte
r-et
hnic
Inte
rvie
w
Shi'i Sunni
Ethnicity
Marginal Effect of an Inter-ethnic Interview90% Confidence Interval
Dependent Variable: Meeting Attendance/Petition Signing
Marginal Effect of X on Y As Z Changes
not significant among Sunni respondents, still it is not far removed from the 90% confidence
interval, and the negative coefficient estimate is a fairly large –0.12 in magnitude. Yet even if
we decide that diffethnic has no effect among Sunni respondents, our difficulty is made only
slightly easier, for we then still face the question of why only Shi‘is would be sensitive to
revealing their political behavior to co-ethnics. The second major problem with this question
sensitivity explanation is that it does not hold up empirically. When we include in our model
of meeting/petition an additional variable to help control for interviewee anxiety,52 the impact
of inter-ethnic interviewing remains and indeed grows even stronger in magnitude among
Shi‘i respondents. This marginal effect we plot above in Figure 5.92. Though the estimate
52
This variable, named refuse, counts the number of times a respondent refuses to answer a select group of the survey’s most sensitive political questions, including many of those investigated already in the political opinion section. (Note that to avoid any selection effect the dependent variable of a given model is not included in this index of questions.) The resulting count is then divided by the total number of sensitive questions, giving a 0 to 1 measure of respondent refusal. In the model referenced here, the refuse variable is highly significant both substantively and statistically (at p = 0.007). The predicted probability of meeting attendance/petition signing for a respondent who refused to answer all the sensitive questions, for example, is just 1%, compared to 41% for one who answered all these questions. Bahrainis who refuse to answer questions, in other words, tend also to lie when they do answer—that, or else they tend to be very outstanding citizens. Cf. supra, note 22.
figure 5.92. Marginal Effect of an Inter-ethnic Interview, with refuse Control
270
among Sunnis is now far from statistical significance, that among Shi‘a gives us additional
confidence that the observed relationship between diffethnic and meeting/petition is not
spurious—or at least is not a consequence of interviewee suspicion of, and misrepresentation
to, co-ethnic interviewers.
We are left, then, with only two basic explanations. Either there remains some form
of systematic bias in Shi‘i responses for which we simply cannot account; or Shi‘i respondents
do in fact tend to exaggerate the extent of their political activism (as measured by meeting/
petition) to Sunni interviewers, even as they tend to moderate their political opinions. This
latter result, if unexpected, is no less interesting, and we shall revisit it shortly.
First, though, we repeat this analysis for the second direct measure of political action
among our Bahraini survey respondents, participation in a demonstration or march.53 If the
previous question may not necessarily imply political action per se—there are many reasons
one may have occasion to “attend a meeting to discuss an issue” or to sign a petition, strictly-
speaking—here the question carries an overt political connotation that could not be lost on
respondents. The distribution of responses, summarized in Figure 5.93 below, would seem
to evidence this fact. Here only 17% of Sunni respondents report partaking in a demonstration
or march in the previous three years, while a full majority of Shi‘a (52% when we exclude
those who refuse to answer or are ostensibly “not sure”) report having done so.
In Table 5.94 below we find the results of our model estimating the determinants of
demonstration participation in Bahrain. Included this time is the additional control variable
refuse, found to be so effective in our previous model (cf. supra, note 52) in capturing respondent
anxiety.54 Looking first at the impact of our standard control variables in the ethnically-
segregated models, we notice that in each case two of the three are related to demonstration.
Female gender as expected is a strong, negative predictor of participation among Sunnis and
Shi‘is alike. Based on the combined model results, the probability of a respondent having
participated in a march or demonstration drops from 43% among male respondents to just
18% among females, all else being equal. Yet this is not the full story. Although the female
control is negatively associated with demonstration among both Sunni and Shi‘i respondents,
this does not mean that females are altogether unlikely to participate. When we calculate the
53
The Arabic, which otherwise follows the wording of the previous question, reads: “ .املشاركة يف مظاهرة أو مسرية ” 54
In the end, however, while the refuse control is once again highly substantively- and statistically-significant, its
inclusion hardly alters our coefficient estimates and does not at all change the substantive interpretation thereof.
271
No Yes Not Sure Refuse
shi'isunni
DemonstrationP
erce
ntag
e of
Res
pond
ents
020
4060
80
predicted probability of demonstration for Shi‘i females based on the Model 3 results, for
example, we find that it is nearly 31%. Among Sunni females, by contrast, the chance is just
8%, according to Model 2. In fact, then, at a 31% probability Shi‘i females are some 63% more
likely to have taken part in a demonstration even than Sunni males, for whom the predicted
probability is 19%, ceteris paribus.
As for the remaining control variables, age is a significant predictor of demonstration
only among Shi‘a, whereas education is only among Sunna. The substantive impact of age
is here again considerable: as per the Model 3 results, the predicted probability that a 20-year-
old Shi‘i respondent will have participated in a demonstration is 60%; that of a 50-year-old,
but 44%. Similarly, the probability that a high school-educated Sunni took part is less than one
figure 5.93. Participation in a Demonstration or March, by Ethnicity
272
table 5.94. The Determinants of Demonstration Participation in Bahrain, estimated three ways
Variables Model 1 Model 2 Model 3
Standard (Probit) Sunnis Only Shi‘is Only
B sb p >|z| B s
b p >|z| B s
b p >|z|
ethnicity -1.777 0.685 0.010 - n/a - - n/a -
diffethnic 0.215 0.218 0.324 -0.114 0.352 0.746 0.206 0.226 0.361
diffeth × eth -0.332 0.402 0.324 - n/a - - n/a -
age -0.00867 0.00587 0.140 0.000822 0.00929 0.930 -0.0138 0.00751 0.066
female -0.737 0.162 0.000 -0.556 0.268 0.038 -0.817 0.198 0.000
education 0.141 0.0661 0.034 0.267 0.108 0.014 0.0654 0.0832 0.432
economy 0.0301 0.147 0.838 0.497 0.236 0.035 -0.00443 0.153 0.977
econ × eth 0.413 0.278 0.137 - n/a - - n/a -
religiosity 0.368 0.208 0.077 0.466 0.296 0.115 0.324 0.215 0.132
relig × eth 0.0432 0.346 0.901 - n/a - - n/a -
refuse 3.120 0.867 0.000 1.759 1.338 0.189 3.700 1.049 0.000 Constant -3.403 0.965 0.000 -5.062 1.544 0.001 -3.289 1.202 0.006
N 382 166 216
Prob. > F 0.0000 0.0382 0.0000
Pseudo R2 0.2363 0.1389 0.1540
Note: All three models report robust standard errors
in ten; a college-educated Sunni, more than one in five. If we exclude female respondents,
these probabilities rise to 14% and 30% of Sunni males, respectively. Among Sunnis, then,
our results suggest that it is the better-educated who incline toward political protest; among
Shi‘is, the youth (and, at least relative to Sunnis, women). Finally, we may note the decisive
impact once again of the refuse variable, whose overall statistical significance is second only
to our gender indicator. Among Shi‘i respondents, for example, those who refuse to answer
one-quarter of the sensitive questions included in the index (that is, when refuse = 0.75), the
273
-.5
-.25
0.2
5
Mar
gina
l Effe
ct o
f Eth
nici
ty
Not Religious Religious
Religiosity of Respondent
Marginal Effect of Ethnicity95% Confidence Interval
Dependent Variable: Meeting Attendance/Petition Signing
Marginal Effect of X on Y As Z Changes
-.25
0.2
5
Mar
gina
l Effe
ct o
f Res
pond
ent R
elig
iosi
ty
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: Demonstration Participation
Marginal Effect of X on Y As Z Changes
figure 5.95. Marginal Effect of Ethnicity, by Respondent Religiosity
figure 5.96. Marginal Effect of Religiosity, by Respondent Ethnicity
274
predicted probability that a respondent answers “Yes” to demonstration is just 26%, compared
to 62% for those who refuse none of the questions. Once more, therefore, either there exists
a strong correlation between distrustful citizens and very upstanding citizens, or else those
who refuse to answer questions also tend to give dishonest answers when they do respond.
As if the picture of Figure 5.93 were not evidence enough, Figure 5.95 confirms that
ethnic group membership is a significant predictor of demonstration across the full range of
its modifying variable religiosity. All else being equal, Sunni ethnicity is estimated to bring
a decreased likelihood of participation of around 32 percentage points among non-religious
individuals and 41 percentage points among religious individuals. (Note that these are absolute
rather than relative changes in probability.) But as this conclusion comes at little surprise,
more notable is the estimated effect of religiosity on Sunni and Shi‘i respondents, depicted
in Figure 5.96. Once more are our expectations confirmed: among Shi‘is, being a religious
person is associated with an absolute change in predicted probability of demonstration of
an estimated 0.14; among Sunnis this marginal effect is 0.05, which if only around one-third
the magnitude of the former is nonetheless statistically-significant.
For a more useful illustration of the influence of religiosity on Bahrainis’ propensity
to take part in demonstrations, consider Figures 5.97 and 5.98 below, which are based on the
combined Model 1 results. From the former, we see that the likelihood that a non-religious
Sunni respondent will have reported taking part in a demonstration is only around 4%,
compared to 10% for a religious Sunni. For Shi‘is, religiosity increases this probability from
an estimated 39% to 54%. Thus, while the absolute change effected among Sunnis is perhaps
smaller in magnitude, the relative influence of religiosity is indeed greater among this group
than among Shi‘a, for whom the baseline likelihood of participation is much higher owing to
the effect of ethnic membership. All in all, being a religious person makes it 38% more likely
that a Shi‘i respondent, and 150% more likely that a Sunni respondent, will have participated
in a demonstration in Bahrain.
When we follow our previous procedure to restrict the sample to male respondents
only, we find that these changes in predicted demonstration owing to religiosity remain
substantively equivalent. The baseline likelihood of participation among Sunnis is now an
estimated 7%, jumping to 15% among the religious. Non-religious Shi‘a males are now 50%
likely to report having participated, compared to 64% among religious individuals. The relative
increases in demonstration probability due to religiosity drop to 25% among Shi‘a and 103%
275
Predicted Petition by Sect
0
10
20
30
40
50
60
70
Sunni Shi‘i
Pre
dict
ed P
roba
bilit
y o
f P
artic
ipat
ion
(%)
Not Religious Religious
Predicted Petition by Sect
0
10
20
30
40
50
60
70
Sunni Shi‘i
Pre
dict
ed P
roba
bilit
y o
f Par
tici
patio
n (%
)
Not Religious Religious
figure 5.97. Predicted Probability of Demonstration Participation
figure 5.98. Predicted Probability of Demonstration Participation, Males Only
276
Household Economy Inter-ethnic Interview
-.25
0.2
5
Mar
gina
l Effe
ct o
f Res
pond
ent E
con
omy
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Economy (+/- 1 STD)90% Confidence Interval
Dependent Variable: Demonstration Participation
Marginal Effect of X on Y As Z Changes
-.
250
.25
Mar
gina
l Effe
ct o
f an
Inte
r-et
hnic
Inte
rvie
w
Shi'i Sunni
Ethnicity
Marginal Effect of an Inter-ethnic Interview90% Confidence Interval
Dependent Variable: Demonstration Participation
Marginal Effect of X on Y As Z Changes
among Sunna. By either measure, then, notwithstanding the mobilizing effect of religiosity
upon Sunni and Shi‘i alike, still ethnic group membership continues to play the most decisive
role in determining individual behavior in the sphere of political action. Even with an
augmenting effect of some 100% to 150%, religious Sunnis remain several times less likely to
have taken part in a political demonstration than their Shi‘i counterparts, religious or not. In
this case, “Sunni or Shi‘i?” still trumps “religious or non-religious?”
The question of “rich or poor Bahraini?,” on the other hand, seems here to apply only
to Sunnis. We observe in Figure 5.99 the marginal effect of a ±1 standard deviation increase
in the economy variable (i.e., a change in the direction of poorer household economy) among
respondents of both ethnic groups. Among Shi‘is the change makes utterly no difference in
predicting a respondent’s probability of demonstration, yet among Sunnis it is associated
with an absolute increase of approximately 0.13. Put in substantive terms, the estimated
likelihood of demonstration participation for a Sunni of “very good” household economy is
7%, all else being equal, of “good” economy 16%, of “poor” economy 29%, and of “very poor”
economy 45%. Among Shi‘is, by contrast, the estimated probability of demonstration increases
from 48% among those who report “very good” economy to 51% among those with “very bad,”
figure 5.99. Marginal Effects of economy and diffethnic
277
a change that in any case is not statistically-distinguishable from 0. Poorer Bahraini Shi‘a, it
turns out, are no more likely to demonstrate than are any other Shi‘a. In rentier language,
when the basis of political conflict is not economics but ethnicity and religion, a wealthier
opposition does not a more mollified opposition make.
We turn our attention, finally, to the other half of Figure 5.99, where the diffethnic
variable is shown to be statistically unrelated to a respondent’s likelihood of demonstration
participation. This result would appear to cast doubt upon our foregoing discussion about
the effects of inter-ethnic interviewing on meeting/petition. Why would respondents tend
to misrepresent the extent of their participation in a relatively innocuous political activity—
signing petitions and attending public inquiries—and yet remain unaffected in their answers
to this even more sensitive question about demonstration? To help resolve this paradox, we
repeat our analysis using an aggregate measure of direct political action that combines these
two variables. This combined indicator is coded 0 only if a respondent reports having never
participated in either a meeting/petition or demonstration—that is, for total non-participation.
A code of 1, then, means a respondent has participated in at least one activity. The principal
benefit of this indicator is that it affords increased statistical leverage with which to estimate
the effect of diffethnic, particularly among Sunni respondents. Whereas the standalone
meeting/petition and demonstration variables had little variation in responses among Sunnis
(with 78% and 83% saying they had not participated, respectively), in our combined indicator
32% of Sunni respondents report having joined in at least one activity. In sum, the issue we
face is not that the estimated effects of diffethnic are inconsistent for these two measures—
in each case this estimate is positive among Shi‘a and negative among Sunnis—but that their
standard errors are so large as to preclude statistical significance. By increasing the variation
in our dependent variable, especially among Sunnis, we should obtain more robust estimates.
The upshot of this reanalysis we see in Figure 5.100.55 Using the aggregate measure of
direct political action, the diffethnic variable proves to be a significant predictor among both
Shi‘is and Sunnis. Equally importantly, its substantive impact remains the same as observed
already in Figures 5.91 and 5.99: Shi‘is who are interviewed by Sunnis tend to exaggerate the
extent of their participation in direct political action, while Sunnis interviewed by Shi‘is tend
55
For the sake of space, we exclude here the full model estimation results, which are otherwise substantively unchanged. Religiosity remains strongly associated with participation among both Sunni and Shi‘i respondents. As before, younger people are more likely to have participated irrespective of ethnicity, while higher education results in more participation only among Sunna. For the influence of household economy, see Figure 5.101.
278
-.25
0.2
5
Mar
gina
l Effe
ct o
f an
Inte
r-et
hnic
Inte
rvie
w
Shi'i Sunni
Ethnicity
Marginal Effect of an Inter-ethnic Interview90% Confidence Interval
Dependent Variable: Demonstration/Meeting/Petition Participation
Marginal Effect of X on Y As Z Changes
-.25
0.2
5
Mar
gina
l Effe
ct o
f Res
pond
ent E
cono
my
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Economy (+/- 1 STD)90% Confidence Interval
Dependent Variable: Demonstration/Meeting/Petition Participation
Marginal Effect of X on Y As Z Changes
figure 5.100. Marginal Effect of diffethnic, using Aggregate Measure of Political Action
figure 5.101. Marginal Effect of economy, using Aggregate Measure of Political Action
279
to understate it. (If one would wonder whether our estimate of the effect of economy on direct
political action might also benefit from this aggregate measure, Figure 5.101 should remove
any doubt. Plainly, it does not.) Having so gained some confidence that the observed effect
of inter-ethnic interviewing is not spurious, we must return to our original question: Why is
the substantive impact of diffethnic on political action opposite that on political opinion? If
Shi‘a tend to moderate their political opinions to Sunni interviewers, and Sunnis tend to give
more anti-government opinions to Shi‘is—of these effects we have ample evidence—then
how can we explain the tendency of Shi‘a to exaggerate their political participation to other-
ethnic interviewers, and of Sunnis to conceal it?
Recall, first of all, that we cannot simply appeal to interviewee anxiety, for which we
duly accounted through use of the refuse indicator. This was seen in Table 5.94 to operate
in the same direction upon Sunnis and Shi‘is. Respondents—all respondents—who refuse to
answer sensitive questions also tend to give more moderate answers when they do respond.
Thus question sensitivity is not a cogent explanation unless we assume arbitrarily that in the
case of questions regarding political actions it is Shi‘i interviewers that influence respondents’
answers, not other-ethnic interviewers. In this case, our interpretation is that Shi‘i fieldworkers
somehow exert the same influence upon respondents from both ethnic groups, making them
underreport the extent of their political actions compared to respondents interviewed by
Sunnis.56 But this raises even more questions than it answers.
To avoid such leaps in argument, we may proffer at once a simpler solution and one
more in keeping with the picture of Bahrain’s ethnic politics constructed over the preceding
chapters: that Sunni and Shi‘i respondents are both merely bluffing. When asked by Sunnis
about their political actions, Shi‘a may tend to exaggerate the extent of their participation to
give the picture of a more vigorous and active opposition; Sunnis, when asked by Shi‘a, to
exaggerate the extent of their non-participation, to signal a more unified and steadfast pro-
government faction. Why this runs counter to our previous findings, wherein inter-ethnic
interviewees consistently give answers closer to those they attribute to their questioner,
requires of course some speculation. Yet we may conjecture that the incentive to misrepresent
an opinion known to be considered deviant—essentially, avoidance of shame—differs from
that to misrepresent an action the revelation of which may convey valuable information to an
adversary. Here the motivation is not to escape shame but the appearance of weakness.
56
Or, equally arbitrarily, we may say that Sunni interviewers make respondents exaggerate their political actions.
280
No Yes Refuse
shi'isunni
Voted
Per
cent
age
of R
espo
nden
ts
020
4060
80
We arrive finally at the last dependent variable in our analysis of political action in
Bahrain, and in our investigation of the Bahrain survey more generally. We have witnessed
already the effects of ethnicity and religiosity on respondents’ views of the 2006 parliamentary
elections. We conclude now by assessing the causes of their participation therein. Asked
directly, “Did you participate in the most recent parliamentary elections that took place on
November 25, 2006?,”57 around 72% of Sunni respondents reported voting, compared to just
58% of Shi‘a. This between-ethnic discrepancy, as well as the relatively high proportion of
refusals—a combined 6% of respondents—reflects once again the controversy surrounding
the vote. The question of electoral participation, recall, precipitated the fissure of al-Wifäq,
with the resulting al-Haqq Movement leading the charge for an electoral boycott. Not to be
outdone, al-Wifäq promulgated a religious directive from Ayatalläh ‘Alï al-Sïstänï in which he
ostensively compelled pious Shi‘is to vote. As it would be the first election with official Shi‘i
57
The Arabic is: “ ؟2006 نوفمرب 25هل شاركت يف آخر انتخابات نيابية جرت يف ”
figure 5.102. Participation in the 2006 Parliamentary Elections, by Ethnicity
281
table 5.103. The Determinants of Electoral Participation in Bahrain, estimated three ways
Variables Model 1 Model 2 Model 3
Standard (Probit) Sunnis Only Shi‘is Only
B sb p >|z| B s
b p >|z| B s
b p >|z|
ethnicity 0.00543 0.571 0.992 - n/a - - n/a -
diffethnic -0.0485 0.202 0.810 0.323 0.313 0.302 -0.0467 0.206 0.820
diffeth × eth 0.208 0.372 0.576 - n/a - - n/a -
age 0.0306 0.00677 0.000 0.0257 0.0112 0.022 0.0342 0.00821 0.000
female -0.0851 0.150 0.569 -0.0261 0.238 0.913 -0.143 0.201 0.478
education 0.104 0.0612 0.090 0.0812 0.0913 0.374 0.138 0.0857 0.107
economy 0.0204 0.147 0.890 0.310 0.196 0.113 0.0455 0.152 0.765
econ × eth 0.262 0.249 0.292 - n/a - - n/a -
religiosity 0.581 0.206 0.005 0.435 0.251 0.083 0.642 0.215 0.003
relig × eth -0.105 0.315 0.740 - n/a - - n/a -
refuse 0.395 0.564 0.484 2.505 0.948 0.008 -0.864 0.817 0.290 Constant -2.102 0.799 0.009 -3.859 1.218 0.002 -1.288 1.063 0.226
N 382 165 217
Prob. > F 0.0001 0.0111 0.0006
Pseudo R2 0.1106 0.1165 0.1016
Note: All three models report robust standard errors
participation (al-Wifäq had boycotted the 2002 vote), Sunnis were urged for their part not to
be lulled into complacency by the prospect that many Shi‘a would reject the elections along
with al-Haqq. The harm of not voting was seen once again to outweigh that of cooptation.
Our model estimation results, summarized in Table 5.103, mirror this history. Sunni
ethnicity is associated with a higher probability of electoral participation among both religious
and irreligious individuals. Its conditional marginal effect, given in Figure 5.104, is an estimated
0.22 among the former category and 0.14 among the latter. Religiosity too is shown to have
282
-.25
0.2
5
Mar
gina
l Effe
ct o
f Eth
nici
ty
Not Religious Religious
Religiosity of Respondent
Marginal Effect of Ethnicity95% Confidence Interval
Dependent Variable: 2006 Election Participation
Marginal Effect of X on Y As Z Changes
-.25
0.2
5
Mar
gina
l Effe
ct o
f Res
pond
ent R
elig
iosi
ty
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Religiosity90% Confidence Interval
Dependent Variable: 2006 Election Participation
Marginal Effect of X on Y As Z Changes
figure 5.104. Marginal Effect of Ethnicity, by Respondent Religiosity
figure 5.105. Marginal Effect of Religiosity, by Ethnicity
283
Predicted Petition by Sect
0
10
20
30
40
50
60
70
80
90
100
Sunni Shi‘i
Pre
dic
ted
Pro
babi
lity
of P
artic
ipat
ion
( %)
Not Religious Religious
a considerable augmenting effect on electoral participation, and this on both Sunni and Shi‘i
respondents, as per Figure 5.105. When we combine these two effects we get Figure 5.106,
which illustrates the substantive influence of both ethnicity and religiosity on participation
in the 2006 elections. As for the first, we see that the influence of ethnicity is such that a
non-religious Sunni is more likely to have voted even than a religious Shi‘i. The relative impact
of religiosity is much stronger among Shi‘a, however, for whom being a religious individual
increases one’s likelihood of participation from 41% to 66%, or by 61%. Among Sunnis, on
the other hand, the baseline probability of voting is 69%, buoyed only to 82% by one’s being
religious. The relative influence of religiosity among Sunnis, then, is an estimated 20%.
Yet how far do our other results accord with the story of the 2006 elections? More
specifically, how important is a respondent’s personal economy in predicting participation? To
be sure, in lobbying their respective constituencies to take part in the vote, political societies
of all stripes appealed constantly to economic well-being. A united Shi‘a bloc in parliament,
argued al-Wifäq, would help raise the standard of living for all ordinary Shi‘is; boycott, on
the other hand, would only ensure that Bahrain’s socio-economic divide will worsen. Sunni
figure 5.106. Predicted Probability of Electoral Participation
284
Household Economy Inter-ethnic Interview
-.25
0.2
5
Mar
gina
l Effe
ct o
f Res
pond
ent E
con
omy
Shi'i Sunni
Ethnicity
Marginal Effect of Respondent Economy (+/- 1 STD)90% Confidence Interval
Dependent Variable: Electoral Participation
Marginal Effect of X on Y As Z Changes
-.
250
.25
Mar
gina
l Effe
ct o
f an
Inte
r-et
hnic
Inte
rvie
w
Shi'i Sunni
Ethnicity
Marginal Effect of an Inter-ethnic Interview90% Confidence Interval
Dependent Variable: Electoral Participation
Marginal Effect of X on Y As Z Changes
representation in parliament, the two Sunni Islamic societies maintained, would guarantee not
only that “miscreants” could not “enact or pass laws incompatible with religious values,” as
al-Ma‘äwdah warned, but also that al-Wifäq could not succeed in directing state benefits to
its Shi‘a constituents at the expense of Sunni citizens. Compared to the effects of religiosity
and ethnicity witnessed already, then, to what extent did these economic-based arguments
convince Bahrainis on both sides to go to the polls?
As we see from the left side of Figure 5.107 above, if they had any effect at all, it was
upon Sunnis only. The estimated marginal effect of economy on the probability of electoral
participation is virtually 0 among Shi‘a respondents and around 10 percentage points among
Sunnis.58 More concretely, the probability that a Sunni of “very good” economy voted is an
estimated 68%, compared to 77% for those of “good” economy, 85% for those of “poor”
economy, and 90% for Sunnis of “very poor” economy. In deciding whether to take part in
the 2006 elections, Sunnis were thus driven simultaneously by ethnic mistrust and by their
58
Though strictly-speaking neither estimate is statistically-significant at the standard level of confidence, if we omit the lowest level of household economy—“very poor”—reported by only 7 (or 2% of) respondents, the effect among Sunnis is then statistically-significant and slightly greater in magnitude at 0.13. We may thus be confident that economy did indeed play a role in driving Sunnis to vote in 2006. Note that even with this omission the effect of economy among Shi‘a is essentially unchanged, at an estimated –0.01.
figure 5.107. Marginal Effects of economy and diffethnic
285
wallets; Shi‘a, it would seem, only by the concerted appeal to their religious conscience made
by al-Wifäq and, eventually, by ‘Alï al-Sïstänï’s fatwä. The complaint of al-Haqq Movement
leaders—that al-Wifäq and its clerical authorities “coerced” Shi‘a to vote—here finds evidence
in the Bahrain survey.59
What else have we learned from our investigation of individual-level political action
in Bahrain? Summarized in Table 5.108 below are the results of all six model estimations,
including those relating to the combined demonstration/meeting/petition indicator. As in
our previous summary of the determinants of political opinion (cf. Table 5.73), here we report
the substantive impact of each of our independent variables of interest on each dependent
variable analyzed. If different in other respects, in one important way this overview mirrors
our previous summary: the results are reassuringly stable across a rather diverse set of models,
some involving dichotomous dependent variables, others not; some measuring direct political
action such as participation in demonstrations, others measuring indirect involvement such
as interest in and following of politics. In every case, our theoretical explanators, in particular
the ethnicity and religiosity variables, exert a similar substantive influence, with higher
religiosity consistently associated with more (or a higher probability of) political activity, and
Sunni ethnicity consistently associated with the opposite save for when that activity itself, as
in the case of electoral participation, is in support of the regime. With this we have all but
confirmed the final theoretical hypothesis underlying our investigation of the Bahrain survey:
Hypothesis 2.2C, which predicts that “Among Sunnis as well as Shi‘is, the strength of ethnic
identity is positively associated with political action.”
As for our wider theoretical prediction that ethnic affiliation has equal or greater
power in explaining individual political action in Bahrain as compared to economic satisfaction,
here our conclusion may be offered with equal confidence. Of the six models summarized in
59
Another way we may come at this question is by looking at respondents’ reported political affiliations. While too many (47% of Shi‘a, including those who answered that “there is no society” that approximates their political views) declined to answer this question to allow a full regression analysis, still the data are instructive. Of those Shi‘a who do identify with a political society, a little over half, 55%, named al-Wifäq. A further 10% of respondents aligned with socialist-leaning Wa‘ad, which attracts Sunnis together with Shi‘is. Around 15% of Shi‘is, finally, mentioned various minor groups, including local charities, human rights organizations, and liberal parties. The remainder—approximately 20%—identified al-Haqq. Now, the average household economy score of those who identify with al-Wifäq is 2.25; of those who name al-Haqq 2.28. A standard, one-tailed difference of means test rejects the hypothesis that the al-Haqq average is larger than that of al-Wifäq supporters, with an associated p-level of 0.5782. This result remains if we exclude those outlying observations at the lowest level of household economy. In sum, Bahrain’s most ardent political opponents, those who refuse to “participate in the political process,” as is said, are not simply its poorest.
286
table 5.108. The Determinants of Political Action among Bahrainis, Summary of Results
Dependent Variable Impact of Independent Variables of Interest
Ethnicityb Religiosity
Economy
e Interviewer
(Shi‘i | Sunni) (Increasing) (Worsening) (Inter-ethnic)
Indirect Actions
interest Less c More| More
a none | none none | none
(7.7%) (7.7%| 5.6%)
news Less c More | More
a none | none none | none
(6.3%) (14.3%| 7.7%) ) Direct Actions
d
meeting/petition Less Likely More | More none | none Moref | none
(45.1%) (64.9% | 59.1%) (50.2%)
demonstration Less Likely More| More none | More none | none
(79.3%) (28.7%| 103.1%) ) (99.5%)
combined Less Likely More | More none | none More | Less
meeting/petition or demonstration (43.1%) (17.1%| 42.0%) (17.7%| 60.0%)
voted More Likely More | More none | More
a none | none
(37.0%) (60.6%| 19.7%) (14.0%)
a Coefficient has the expected sign and is significant in magnitude, but its related p-value falls slightly outside the p < 0.100
threshold of significance b Recall that since our ethnicity variable is coded 1 for Sunnis, the effect of ethnic group membership is always expressed
as the effect of a respondent’s being a Sunni rather than a Shi‘i. This choice is arbitrary of course, and to express the effect in terms of Shi‘i ethnicity one merely has to reverse the sign of the coefficient. c Ethnicity’s effect is only statistically-significant among religious individuals.
d Recall that because each of these “direct action” measures is dichotomous, the reported percentages represent relative
changes in probability that a respondent answers “Yes.” To offer the most conservative estimates possible, except in the case of voted (where gender does not influence the likelihood of participation) the probabilities are calculated here for
male respondents only. Cf., e.g., Figures 5.90 and 5.98. e The percentages reported here correspond to the percent change in likelihood of participation associated with a change
in respondent economy from –1 standard deviation to +1 standard deviation, or from approximately 1.5 to 2.7 on our 1-4
scale. So as to remain directly comparable to the impact of religiosity, this probability is also calculated only for men in the case of the dependent variable demonstration. f Based on the revised model including the refuse control variable
Table 5.108, household economy is a significant predictor of respondent behavior in just two
cases, and in both of these only among Sunni ethnics. Moreover, in each case the substantive
impact of economy not only is less than or on par with that of ethnic group membership, but
it is outweighed even by the additional, augmenting effect of religiosity. By any standard,
the explanatory power of economic well-being simply cannot compare to that of ethnicity
and personal religiosity. In today’s Bahrain, “Rich or poor?” is a secondary matter beside the
more important questions: “Sunni or Shi‘i?”—then: “How Sunni?” and “How Shi‘i?”
287
More notable than the effect of economy on political action per se, then, is the fact
that it this effect operates upon Sunnis only, particularly as this represents a reversal of the
trend witnessed in the realm of political opinion. There, in four out of six models the economy
variable proved a significant predictor of political opinion among Shi‘a. Among Sunnis, it was
associated with respondents’ answers only twice. And so the overall picture that emerges
regarding the substantive influence of economic well-being is seemingly paradoxical: Shi‘a of
more depressed economy tend to form more negative opinions of the government, but are no
more likely to act out as a result; while poorer Sunnis are more likely to take part in direct
political action yet seem to be relatively unaffected in their political opinions, at least as
compared to the Shi‘a. But such is Bahrain. Sunnis remain ideologically-supportive of the
government qua protector of the status quo even as they register their political grievances
about economic conditions. Shi‘a remain opposed to the political status quo on principle, a
position perhaps bolstered by but in the end independent of material circumstances.
Conclusion: Rentier State Theory versus Rentier State Reality in Bahrain
Utilizing the individual-level data only now available from my 2009 mass survey, the present
chapter sought empirical evidence of the patent disconnect between rentier state theory and
rentier state reality in Bahrain as demonstrated in earlier pages. In this it has succeeded. Our
analysis has shown that neither of the two basic arguments forwarded to explain the supposed
link between rentier wealth and political quiescence obtains in the case of Bahrain. Individuals
are bound to the government neither by the promise of public-sector employment nor by
freedom from taxation. Instead, Bahrainis’ orientations toward the state—as represented both
by their opinions and actions—are determined above all by their position in the larger ethnic
conflict that divides society. Sunnis tend to align with the prevailing regime, Shi‘a tend to
oppose it, and individuals from both groups are only further entrenched in their respective
positions as their ethno-religious identification increases. At the same time that it further
divides opinion, finally, the latter serves also to augment political activity among all citizens,
begetting a society that is neither politically-quiescent nor satisfied economically—begetting,
in short, an ethnically-divided, dysfunctional rentier state.
We saw first how an individual is substantially more likely—nearly 40% more likely—
to be employed in Bahrain’s public sector if s/he is a Sunni ethnic, even after accounting for
288
other relevant individual-level factors. Beblawi’s theoretical conception of state employment
as a “legitimate aspiration” for all citizens of allocative economies was shown accordingly to
depend on naïve assumptions about the bases of and preconditions for government service in
rentier states. In Bahrain and elsewhere in the Arab Gulf, public-sector employment does not
lead to political allegiance; political allegiance leads to public-sector employment, and even
more so when the employment in question carries national security implications. This
interpretation was bolstered by our finding that not a single Shi‘i of the 117 employed males
interviewed reported working for the police or armed forces. By contrast, 17 separate Sunni
households, representing some 13% of all those surveyed, reported at least one member
working in the police or military. In sum, Bahrain’s leaders cannot effectively use
employment to appease would-be opponents precisely because they disproportionately reject
as unsuitable those whom they would most like to appease.
Not only are Bahraini Shi‘a less likely to find work in the public sector, moreover, they
also tend to hold lower-level professional positions when they are employed, and this in the
state and non-state sectors equally. An individual of Sunni ethnicity, we found, is predicted
in each case to occupy a professional level that is around 15% higher on our 11-point scale.
In the private sector, this effect is substantively equivalent to more than a 10-year advantage
in age. Its relative influence in the public sector is less clear, however, because professional
level in the state sector was shown to depend on fundamentally different criteria: unlike in
the private sector, here females tend to occupy higher positions, for example, while one’s age
plays no role in determining professional level. In sum, contrary to the notion of the pubic
sector of rentier economies as a sort of parallel job market available to those less able to find
work in private industry, our empirical findings suggest rather that both employment and
advancement proceed in Bahrain not on the basis of economic need but political expediency,
personal favors, and of course, whether one happens to be a Sunni or a Shi‘i.
The second half of our survey analysis investigated the relative influences of economic
well-being and ethno-religious identity in determining Bahrainis’ political views and behavior.
We found that the political opinions of ordinary Bahrainis—from their trust in important
state institutions, to their overall satisfaction with government performance, to their views
of Bahrain’s 2006 parliamentary election—are determined almost entirely along ethnic lines,
with Sunni ethnicity associated with much more pro-government opinion even after one
accounts for the effects of age, gender, and education. What is more, heightened religiosity
289
among respondents serves only to augment this between-group difference, driving opinion
farther apart. Among Sunni ethnics, measures of personal religiousness correspond to even
more favorable government opinion; among Bahrain’s Shi‘a, to more anti-government views.
At the same time that religion pushes Shi‘is toward more adversarial political orientations,
that is, it marshals Sunnis further to the regime’s defense.
As for the influence of household economy, on the other hand, its effect on opinion
is inconsistent and relatively weak. Where it is shown to alter Bahrainis’ views, its substantive
impact remains on average some two to three times less in magnitude than that of ethnic
affiliation. Even the additional, augmenting effect of personal religiosity is more robust. While
it is true then that, on the margin, more economically dissatisfied Bahrainis tend to hold less
favorable positions toward their government, this relationship is but a footnote in the larger
narrative of Bahraini politics, which has been woven firmly around ethnic difference since
the time the Äl Khalïfa and their Sunni tribal allies arrived on the island. In none of our six
models of political opinion does the influence of economic satisfaction negate that of ethnic
group membership, much less the combined impact of ethnicity and religiosity. That untaxed
and economically-satisfied citizens make pro-government citizens is a notion ill-suited to the
socio-political conditions that, unfortunately, prevail today in much of the Arab Gulf.
Neither, according to our results, do materially better-off citizens necessarily make
more inactive citizens. In only two of our six models of political action is household economy
a significant predictor of direct or indirect participation, and there only among Bahrain’s
Sunnis. Shi‘a citizens protest, sign petitions, attend public inquiries, and vote in elections
not on the basis of economy, not because they seek redress for economic grievances, but on
principle. Their political engagement stems not from material dissatisfaction but from
dissatisfaction with the regime as a whole, wherein they find themselves limited as a group
in political power and social standing on the basis of ethnicity. Only among Sunnis, then, do
we have evidence that better economy elicits more political quiet. This also implies on the
other hand that Bahrain’s rulers do not earn a free pass from Sunni citizens merely on account
of shared ethnicity. For their near-unwavering support, and for their help in keeping the
government’s fiercest critics at bay, ordinary Sunnis expect something in return.
Yet even among Sunnis the role of economy in determining one’s propensity for
political action is only half the story, or indeed is less. More powerful than this is the impact
290
once again of ethno-religious identification. Although as a group Sunnis are much less likely
to engage in political action, personal religiosity augments this likelihood among individual
Sunnis and Shi‘is alike, and this in every one of the six models we analyzed. Thus proceeds
the battle over Bahrain’s political status quo, pushed on the one side by Shi‘a reformists, pulled
on the other by a countervailing force of Sunnis acting, to quote al-Ma‘äwdah a final time,
“to counter probable harm.”
The survey offered still one more insight into ethnic politics in Bahrain. A last but
perhaps no less revealing result of our investigation is the effect of inter-ethnic interviews on
individuals’ survey responses. In short, Bahrainis are unwilling to divulge their true political
positions and behavior to members of the other group. Both Sunni and Shi‘i respondents
misrepresent themselves so as to appear more in step with the opinions they ascribe, purely
on the basis of ethnicity, to their interviewer. Shi‘i respondents misstate their views to appear
much more pro-government, more Sunni-like, whereas Sunnis do precisely the opposite,
offering answers that are more anti-government, more Shi‘i-like. If one would look only at
inter-ethnic interviewees, one must conclude that it is the Shi‘a who are the champions of
the Bahraini regime, the Sunnis its detractors. In the realm of political action, these effects
are reversed: now Shi‘is tend to exaggerate the extent of their participation in order to convey a
stronger and more active opposition; Sunnis, when asked by Shi‘a, exaggerate the extent of
their non-participation, to signal a more resolute pro-government bloc. Among the many
dysfunctions contributing to the dysfunctional rentier state that is Bahrain, then, this deep-
seated ethnic mistrust—combined, it seems, with political posturing—is not the least telling.
291
Notes for Chapter 5
Brambor, Thomas, William Roberts Clark, and Matt Golder. 2006. “Understanding Interaction Models: Improving Empirical Analyses.” Political Analysis 14(1): 63-82.
Bushway, Shawn, Brian D. Johnson, and Lee Ann Slocum. 2007. “Is the Magic Still There? The
Use of the Heckman Two-Step Correction for Selection Bias in Criminology.” Journal of
Quantitative Criminology 23(2): 151-178. Heckman, James J. 1976. “The Common Structure of Statistical Models of Truncation, Sample
Selection, and Limited Dependent Variables and a Simple Estimator for Such Models.” Annals of Economic and Social Measurement 5(4): 475-492.
Sigelman, Lee and Langche Zeng. 1999. “Analyzing Censored and Sample-Selected Data with
Tobit and Heckit Models.” Political Analysis 8(2): 167-182. Vance, Colin. 2006. “Marginal Effects and Significance Testing with Heckman’s Sample
Selection Model: A Methodological Note.” RWI Discussion Papers N° 39. RWI Essen. Available at: <http://repec.rwi-essen.de/files/DP_06_039.pdf>.